CN111078832A - Auxiliary response method and system for intelligent customer service - Google Patents
Auxiliary response method and system for intelligent customer service Download PDFInfo
- Publication number
- CN111078832A CN111078832A CN201911087804.9A CN201911087804A CN111078832A CN 111078832 A CN111078832 A CN 111078832A CN 201911087804 A CN201911087804 A CN 201911087804A CN 111078832 A CN111078832 A CN 111078832A
- Authority
- CN
- China
- Prior art keywords
- text
- vertex
- response
- word
- keywords
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000004044 response Effects 0.000 title claims abstract description 197
- 238000000034 method Methods 0.000 title claims abstract description 72
- 238000000605 extraction Methods 0.000 claims abstract description 58
- BUAJNGPDPGKBGV-UHFFFAOYSA-N 1-(1-phenylcyclohexyl)piperidin-1-ium;chloride Chemical compound [Cl-].C1CCCC[NH+]1C1(C=2C=CC=CC=2)CCCCC1 BUAJNGPDPGKBGV-UHFFFAOYSA-N 0.000 claims description 29
- 239000013598 vector Substances 0.000 claims description 26
- 238000004422 calculation algorithm Methods 0.000 claims description 21
- 239000011159 matrix material Substances 0.000 claims description 15
- 230000011218 segmentation Effects 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 4
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 238000004140 cleaning Methods 0.000 description 14
- 238000011156 evaluation Methods 0.000 description 14
- 238000012423 maintenance Methods 0.000 description 14
- 238000010586 diagram Methods 0.000 description 10
- 230000008569 process Effects 0.000 description 9
- 230000006870 function Effects 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 238000009434 installation Methods 0.000 description 5
- 238000005406 washing Methods 0.000 description 5
- 239000000284 extract Substances 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- FFBHFFJDDLITSX-UHFFFAOYSA-N benzyl N-[2-hydroxy-4-(3-oxomorpholin-4-yl)phenyl]carbamate Chemical compound OC1=C(NC(=O)OCC2=CC=CC=C2)C=CC(=C1)N1CCOCC1=O FFBHFFJDDLITSX-UHFFFAOYSA-N 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- -1 inter-word relation Substances 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000013138 pruning Methods 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
- 230000007723 transport mechanism Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/3332—Query translation
- G06F16/3334—Selection or weighting of terms from queries, including natural language queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/01—Customer relationship services
- G06Q30/015—Providing customer assistance, e.g. assisting a customer within a business location or via helpdesk
- G06Q30/016—After-sales
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- Human Computer Interaction (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Finance (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an auxiliary response method and an auxiliary response system for intelligent customer service, wherein the method comprises the following steps: acquiring user consultation information, and extracting keywords from the response text based on the interword relationship; positioning response knowledge matched with the user consultation information in a response text based on the keywords; and feeding back the response knowledge to the customer service staff so that the customer service staff can respond to the user consultation information according to the response knowledge. The auxiliary response method and the auxiliary response system of the intelligent customer service are used for assisting the customer service staff to make a response, improving the extraction accuracy of the keywords and assisting the customer service staff to improve the working efficiency.
Description
Technical Field
The invention relates to the field of intelligent household appliances, in particular to an auxiliary response method and an auxiliary response system for intelligent customer service.
Background
The customer service seat personnel usually adopt the knowledge basis as product use instruction when answering the user consultation. The system usually extracts keywords from the text of the product use instruction, and the seat personnel can search the keywords on the customer service platform according to the input of the user, so that the related knowledge in the instruction is positioned, and the customer service response is completed.
At present, the performance of the system keyword extraction result and the search keyword extracted by the seat personnel according to the inquiry of the user directly influence the knowledge query performance. Keyword extraction refers to an automated technique for automatically extracting words or phrases with importance and theme from text through a computer program.
However, the conventional keyword extraction method is mainly a supervised method. The keyword extraction accuracy of the supervision method is high, but a large amount of corpus information needs to be labeled manually, and a large amount of labor time is needed.
Disclosure of Invention
In a first aspect, the present application provides an auxiliary response method for intelligent customer service, which is used for assisting a customer service person to make a response, and the method includes:
acquiring user consultation information, and extracting keywords from the response text based on the interword relationship;
positioning response knowledge matched with the user consultation information in the response text based on the keyword;
and feeding back the response knowledge to customer service staff so that the customer service staff can respond to the user consultation information according to the response knowledge.
In a second aspect, the present application provides an auxiliary response system for intelligent customer service, which is used for assisting a customer service person to make a response, and includes:
the extraction module is used for acquiring the user consultation information and extracting the key words from the response text based on the inter-word relation;
the matching module is used for positioning response knowledge matched with the user consultation information in the response text based on the keywords;
and the feedback module is used for feeding the response knowledge back to customer service staff so that the customer service staff can respond to the user consultation information according to the response knowledge.
Compared with the prior art, the auxiliary response method and the auxiliary response system for the intelligent customer service provided by at least one embodiment of the application have the following beneficial effects: the method is used for assisting customer service personnel to make a response, firstly, keyword extraction of a response text is carried out based on an interword relation, keyword extraction is optimized, keyword extraction accuracy is improved, and the technical problems that an existing keyword automatic extraction scheme is poor in evaluation index, low in performance and the like are solved. Secondly, the response knowledge in the response text determined based on the extracted keywords can be used as the reference of the customer service staff to assist the customer service staff in responding, so that the customer service staff can quickly, accurately and comprehensively respond to the consultation of the user, and further assist the customer service staff in improving the working efficiency.
In some embodiments of the present invention, the following effects can also be achieved:
1. the local similarity of the text is adopted to measure the similarity between words in the application text, so that the extraction of the keywords based on the semantic relation between the words is realized, the extraction accuracy of the keywords is improved, and the problem that the extraction accuracy of the keywords is reduced because the semantic relation between the words cannot be reflected only by adopting a TF-IDF method or a TextRank method in the existing scheme can be solved.
2. Constructing an uncertain graph model of the response text by calculating the local similarity of the word vector, and taking each word in the response text as a vertex; adding connecting edges between all the vertexes; and taking the local similarity of the text between the words represented by the two vertexes as the probability of the existence of the connecting edge of the two vertexes. The keyword indexes in the response text are calculated based on the uncertain graph, so that an optimized keyword extraction performance evaluation index is provided, and the technical problems of poor evaluation index, low performance and the like of the conventional keyword automatic extraction scheme are solved.
In some embodiments of the present invention, the keyword indicator in the response text is calculated based on the uncertainty map, and the following effects can be achieved:
1. and calculating the vertex density of each vertex according to the uncertain graph, and using the vertex density as a keyword evaluation index to evaluate the extraction quality of the keywords in the uncertain graph.
2. Calculating vertex density of each vertex according to the uncertainty map, and determining frequency value IDF (w) of each word in the response text based on TF-IDFi) And determining a score TRank (w) of each word in the answer text based on the TextRank algorithmi) The vertex density and frequency value IDF (w)i) And score value TRank (w)i) And combining the plurality of characteristics to serve as a keyword evaluation index so as to evaluate the extraction quality of the keywords in the uncertain graph.
In some embodiments of the present invention, the vertex density of each vertex is calculated according to the uncertainty map, and the following effects can be achieved:
1. and adopting a candidate keyword extraction algorithm under a certain step length, wherein the algorithm calculates the vertex density of the corresponding vertex of each word under the certain step length. And calculating the vertex density based on the uncertain graph to obtain the weight of the keywords in the uncertain graph so as to evaluate the extraction quality of the keywords in the uncertain graph.
2. And adopting a self-adaptive candidate keyword extraction algorithm, wherein the algorithm obtains the vertex density of the corresponding vertex of each word through step length self-adaptive automatic calculation. And carrying out iterative calculation of the vertex density based on the uncertain graph to obtain the weight of the keywords in the uncertain graph so as to evaluate the extraction quality of the keywords in the uncertain graph.
In some embodiments of the present invention, the local similarity of the text is used to measure the similarity between words in the application text, and the following effects can be achieved:
1. according to the similarity between two word vectors in the text and the text interval between two words, the method adopts
Calculating local similarity LocalSim (w) of texti，wj) The method and the device realize extraction of the keywords based on the semantic relation among the words and improve the extraction accuracy of the keywords.
2. The space distance between each word in the text is measured by adopting the text space to determine the local similarity of the text, so that the extraction of the keywords based on the semantic relation between the words is realized, and the extraction accuracy of the keywords is improved.
Additional features and advantages of the application will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by the practice of the application. Other advantages of the present application may be realized and attained by the instrumentalities and combinations particularly pointed out in the specification and the drawings.
Drawings
The accompanying drawings are included to provide an understanding of the present disclosure and are incorporated in and constitute a part of this specification, illustrate embodiments of the disclosure and together with the examples serve to explain the principles of the disclosure and not to limit the disclosure.
Fig. 1 is a flowchart of an auxiliary response method of an intelligent customer service according to an embodiment of the present invention;
FIG. 2 is an exemplary diagram of a product specification provided by an embodiment of the present invention;
fig. 3 is a flowchart of extracting keywords from a response text based on an interword relationship according to an embodiment of the present invention;
FIG. 4 is a block diagram of a system for extracting keywords from response text according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a CBOW model provided in the present embodiment;
fig. 6 is a schematic structural diagram of a 4-step uncertainty map provided in an embodiment of the present invention;
fig. 7 is a schematic structural diagram of an auxiliary response system of an intelligent customer service according to an embodiment of the present invention.
Detailed Description
The present application describes embodiments, but the description is illustrative rather than limiting and it will be apparent to those of ordinary skill in the art that many more embodiments and implementations are possible within the scope of the embodiments described herein. Although many possible combinations of features are shown in the drawings and discussed in the detailed description, many other combinations of the disclosed features are possible. Any feature or element of any embodiment may be used in combination with or instead of any other feature or element in any other embodiment, unless expressly limited otherwise.
The present application includes and contemplates combinations of features and elements known to those of ordinary skill in the art. The embodiments, features and elements disclosed in this application may also be combined with any conventional features or elements to form a unique inventive concept as defined by the claims. Any feature or element of any embodiment may also be combined with features or elements from other inventive aspects to form yet another unique inventive aspect, as defined by the claims. Thus, it should be understood that any of the features shown and/or discussed in this application may be implemented alone or in any suitable combination. Accordingly, the embodiments are not limited except as by the appended claims and their equivalents. Furthermore, various modifications and changes may be made within the scope of the appended claims.
Further, in describing representative embodiments, the specification may have presented the method and/or process as a particular sequence of steps. However, to the extent that the method or process does not rely on the particular order of steps set forth herein, the method or process should not be limited to the particular sequence of steps described. Other orders of steps are possible as will be understood by those of ordinary skill in the art. Therefore, the particular order of the steps set forth in the specification should not be construed as limitations on the claims. Further, the claims directed to the method and/or process should not be limited to the performance of their steps in the order written, and one skilled in the art can readily appreciate that the sequences may be varied and still remain within the spirit and scope of the embodiments of the present application.
Keyword extraction refers to an automated technique for automatically extracting words or phrases of importance and subject from text by a computer program. The conventional keyword extraction method can be classified into a supervised method and an unsupervised method. The keyword extraction accuracy of the supervision method is high, but a large amount of corpus information needs to be labeled manually, a large amount of labor time is needed, and meanwhile, the quality of the corpus directly influences the accuracy of the model. Therefore, the method commonly used in the industry is an unsupervised method.
The conventional unsupervised method employs techniques including statistical methods, theme-based methods, network graph-based methods, and the like. For example, the Term Frequency-Inverse Document Frequency algorithm (Term Frequency-Inverse Document Frequency, abbreviated as TF-IDF) is a statistical method, and is based on the assumption that: if a word appears in a text with a high frequency TF and rarely appears in other articles, the word or phrase is considered to have a good classification capability and is suitable for classification. The method is simple and rapid, but in essence, IDF is weighting which tries to suppress noise, and in addition, words with small text frequency are considered to be important simply, words with large text frequency are useless, in many short texts, the assumption is not completely correct, the word frequencies of many words are close, so that TF terms cannot play a role, and meanwhile, the method cannot reflect the semantic relation among the words. Other improved TF-IDF takes the assumption as a basic characteristic, and combines other characteristics such as part of speech, but rarely considers the semantic relation among words, so that the keyword extraction effect is poor.
The text sorting algorithm (TextRank) is a commonly used classic network graph-based keyword extraction algorithm. The method is based on two assumptions: if a word appears after many words, it is important to say that the word is important; a word is followed by an important word, and the importance of the word also increases. The method comprises the steps of taking each word as a point in a graph, calculating the score of each word by adopting a random walk method, and judging the key degree of the word according to the level of the score. However, the method for constructing the network graph through the co-occurrence frequency often forms a chain graph structure when aiming at short texts, so that the accuracy is reduced, and meanwhile, the TextRank cannot reflect the semantic relation between words.
Compared with the method, the invention provides a text local similarity formula among words by combining a related model word2vec for generating word vectors and a special structure of a product specification, establishes an uncertain graph model through an inter-word relation, provides a vertex density concept and a candidate keyword evaluation index DEN by referring to a graph clustering related method, and provides a candidate keyword extraction algorithm based on an uncertain graph. And a keyword evaluation optimization standard with various characteristics can be provided by combining IDF and TextRank, the specific process does not depend on external manual labeling data, the characteristics of word frequency, word vectors, inter-word relation, product instruction book structure and the like are comprehensively utilized, and the performance of keyword extraction is obviously improved.
Fig. 1 is a flowchart of an auxiliary response method for intelligent customer service provided in an embodiment of the present invention, and as shown in fig. 1, the auxiliary response method for intelligent customer service provided in an embodiment of the present invention may be used for assisting a customer service person to make a response, and specifically may include the following steps:
s101: and acquiring user consultation information, and extracting keywords from the response text based on the interword relation.
The auxiliary response method for the intelligent customer service provided by the embodiment can be suitable for a customer service system of an intelligent household appliance, a customer service system of online shopping, a customer service system of the communication industry and the like, and can provide corresponding response information for consultation information provided by a client so that a customer service worker can respond to the client according to the response information.
The execution main body of the embodiment may be an auxiliary response system of the intelligent customer service, which may be an independent device, or may be an intelligent terminal, and the intelligent terminal may be, but is not limited to, an application program in the intelligent terminal or an installation-free application program in the terminal. The installation-free application program is an application program which is embedded in a third-party application program and can be used without being downloaded and installed by a user, such as a WeChat applet (abbreviated as an applet), a light application, a headline number or a service number.
The user consultation information refers to a question which is provided by a user to the intelligent customer service and needs to be solved by the intelligent customer service, and the user consultation information can be but is not limited to product related information (such as product model, function or price) and after-sale consultation information (such as refund, product installation or product maintenance).
The user consultation information can be text information input by the user through an input key or an input box, and can also be voice information input by voice.
In this embodiment, after the auxiliary response system of the intelligent customer service obtains the user consultation information, the keywords are extracted from the response text based on the inter-word relationship. The response text refers to text information related to the user consultation information, and the response text may include, but is not limited to, a product specification (mainly including relevant model, function or maintenance of a product), a product operation manual (mainly including installation or maintenance of a product), a product ordering information table (mainly including ordering price and relevant performance parameters of a product), or an after-market service manual (mainly including after-market terms or after-market solutions).
The response text may be a text pre-stored in an auxiliary response system of the intelligent customer service, a text acquired by a third-party server (such as a cloud server), or a text acquired by a web page link. When the response text is pre-stored, a plurality of texts can be pre-stored in the auxiliary response system of the intelligent customer service, and each text corresponds to one type of user consultation.
Optionally, in this embodiment, when obtaining the user query information, the response text related to the user query information may be searched according to specific information content (for example, a keyword) of the query information or an information type (for example, a function query, a price query, an installation query, or the like) of the query information. For example, when the user consultation information includes that the price or the type belongs to the price consultation, the relevant response text is determined to be a product ordering information table; when the user consultation information comprises refund or the type belongs to refund consultation, determining that the related response text is an after-sales service manual; when the user consultation information comprises installation or the type belongs to installation consultation, determining that the related response text is a product operation manual; when the user consultation information comprises maintenance or the type belongs to maintenance consultation, the relevant response text is determined to be a product specification. Fig. 2 is an exemplary diagram of a product specification provided by an embodiment of the present invention, which may include specific steps and operational instructions for cleaning and maintenance, as shown in fig. 2.
In this embodiment, when extracting the keywords from the response text based on the inter-word relationship, the corresponding keywords may be extracted from the entire response text based on the inter-word relationship, or the response text may be split in advance according to the structure of the response text, for example, the response text may be split by chapters or bars, and the keywords may be extracted from each split portion based on the inter-word relationship.
Optionally, according to the structure of the response text, when the response text is split in advance, a general word (or called a heading word) may be set for each split part, such as a chapter general word in fig. 2: chapter five cleaning and maintenance. When searching for the corresponding response text based on the user consultation information, a specific certain split part (such as the fifth chapter cleaning and maintenance in fig. 2) of the corresponding text can be found according to the specific information content (such as keyword maintenance) of the consultation information or the information type (such as maintenance query) of the consultation information, and the keywords are extracted from the split part, so that the extraction of the keywords is optimized based on the structural characteristics of the response text, and the extraction accuracy of the keywords is improved.
S102: and positioning the response knowledge matched with the user consultation information in the response text based on the keywords.
In this embodiment, after determining the keywords of the response text based on the inter-word relationship, the relevant knowledge in the response text is located based on the extracted keywords, where the relevant knowledge refers to response knowledge matched with the user consultation information to assist the customer service staff to answer.
Specifically, the response knowledge located in the response text based on the keywords and matched with the user consultation information may include the following two implementation manners:
the first implementation mode comprises the following steps: consulting keywords in the user consulting information can be determined based on TF-IDF, TextRank or dependency syntactic analysis (DP for short), the consulting keywords are matched with keywords extracted from the response text, keywords with the same or similar semantics as the consulting keywords are found out, and the positions of the keywords in the response text are located.
For example, a response text base is preset in the system, if it is determined that the query keywords in the user query information include maintenance and cleaning, based on keyword maintenance, the response text can be found from the response text base as the product specification or the fifth chapter of the product specification shown in fig. 2, and then the keywords are extracted from the product specification or the fifth chapter of fig. 2, and if the extracted keywords can include cleaning, washing and storing, at this time, the part of the specification or the fifth chapter of fig. 2 having the cleaning words can be located, and because the washing and cleaning semantics are similar, the part of the specification or the fifth chapter of fig. 2 having the cleaning and washing words can also be located.
The second implementation mode comprises the following steps: the user intention in the user consultation information can be determined based on a Convolutional Neural Network (CNN), the user intention is matched with the keywords extracted from the response text, keywords with the same or similar types as the user intention are found, and the position of the keyword in the response text is located.
Wherein each user advisory information may contain one or more user intentions, each of which may perform a certain specified function.
For example, a response text library is preset in the system, and if it is determined that the user intention in the user consultation information is: i.e. the type of intention is maintenance and cleaning. First, based on the intention type maintenance and cleaning, the answer text which is the product specification or the fifth chapter of the product specification shown in fig. 2 can be searched from the answer text library, then the keywords are extracted from the product specification or the fifth chapter of fig. 2, and if the extracted keywords can comprise cleaning, washing and storing, the part with the cleaning words in the specification or the fifth chapter of fig. 2 can be located, and the part with the cleaning words in the specification or the fifth chapter of fig. 2 can also be located because the washing semantics are similar to the cleaning semantics.
The relevant knowledge located in the response text based on keyword matching may also adopt other keyword matching and locating methods in the prior art, which is not limited and described herein in this embodiment.
S103: and feeding back the response knowledge to the customer service staff so that the customer service staff can respond to the user consultation information according to the response knowledge.
In this embodiment, the response knowledge in the response text determined based on the extracted keywords may be used as a reference for the customer service staff to assist the customer service staff in responding, so that the customer service staff can quickly, accurately and comprehensively respond to the user's consultation.
The auxiliary response method of the intelligent customer service is used for assisting the customer service staff to make a response, firstly, the keywords of the response text are extracted based on the interword relation, the extraction of the keywords is optimized, the extraction accuracy of the keywords is improved, and the technical problems that an existing automatic keyword extraction scheme is poor in evaluation index, low in performance and the like are solved. Secondly, the response knowledge in the response text determined based on the extracted keywords can be used as the reference of the customer service staff to assist the customer service staff in responding, so that the customer service staff can quickly, accurately and comprehensively respond to the consultation of the user, and further assist the customer service staff in improving the working efficiency.
Further, in the above-described embodiment, the inter-word relationship represents a local similarity of text between words. The local similarity of the text between the words can be determined according to the similarity between two word vectors in the text and the text interval between the two words.
Specifically, fig. 3 is a flowchart of extracting a keyword from a response text based on an inter-word relationship according to an embodiment of the present invention, and fig. 4 is a system framework diagram of extracting a keyword from a response text according to an embodiment of the present invention, as shown in fig. 3 and fig. 4, the extracting a keyword from a response text based on an inter-word relationship according to an embodiment of the present invention may include:
s301: and performing word segmentation and word vectorization on the response text, and determining the local similarity of the text between any two words in the response text.
In this embodiment, by segmenting words of the response text or the split part of the response text, determining a word vector of each word, and determining an inter-word semantic relationship (i.e., local similarity of the text) based on the word vector similarity of any two words in the response text, an optimized keyword extraction performance evaluation index is provided, and a system and method for automatically extracting keywords are provided, so that the technical problems of poor evaluation index, low performance and the like of the existing keyword automatic extraction scheme are solved.
The purpose of word vectorization is to digitize each word in the response text into a vector of fixed length, which was originally proposed by Hinton, and words can be mapped into a low-dimensional and dense real number vector space, so that words with similar word senses have similar distances in space.
The training of the word vector is carried out while training the language model, and the word vector is obtained incidentally. The formalized description of the language model is given by a string, and the probability P (w) that it is natural language is calculated1，w2，...，wn)，wiDenotes any word in the application text, i ═ 1,2 … … n.
Optionally, fig. 5 is a schematic structural diagram of the CBOW Model provided in this embodiment, and as shown in fig. 5, the word vector calculation method in this embodiment may use a Continuous Bag of Words Model (CBOW Model for short) of word2vec for calculation. Wherein word2vec is a correlation model for generating word vectors, the training input of the CBOW model is a word vector corresponding to a word related to the context of a certain feature word, and the output is the word vector of the specific word. The implementation principle of the word2vec and CBOW models is the same as that of the prior art, and the description of this embodiment is omitted here.
In this embodiment, for short texts in a response text (such as a product specification), when extracting keywords, the text itself has no beginning and end paragraphs, and the importance degree of the beginning and end sentences is not obviously different from other sentences. In fact, in the product use specification, when a section of text is converted into an ordered phrase, the keywords generally do not appear at the head and tail positions; meanwhile, the keywords can express the subject meaning of the text, and the semantics of the keywords and other words in the text are similar under the general condition. If word vector similarity is used to represent the degree of semantic similarity between words, a keyword of a short text will have the following two features:
(1) the position of the keyword is generally not located at the beginning and the end of the text and may appear in the text for many times;
(2) the similarity of the word vectors of the keywords and other words in the text is high.
From these two features, word-to-word similarity can be constructed.
Optionally, performing word segmentation and word vectorization on the response text, and determining local similarity of the text between any two words in the response text, may include:
performing word segmentation and stop word removal on the response text to obtain an ordered word group W with the length of n ═ W1，w2，w3，...，wn}; determining two words w based on a word vectoriAnd wjSimilarity between vectors sim (w)i，wj) (ii) a Determining two words wiAnd wjText interval tDis (w) in betweeni，wj) (ii) a According to sim (w)i，wj) And tDis (w)i，wj) Determination of wiAnd wjLocal similarity LocalSim (w) of text therebetweeni，wj)。
Wherein,
tDis(wi，wj) 1+ a, a denotes two words wiAnd wjThe number of words in the interval between.
In this embodiment, the process of segmenting a text to remove stop words is referred to as a preprocessing process. Specifically, an ordered phrase W with a length n can be obtained from a text through a preprocessing process1，w2，w2，...，wnIn this embodiment, the text interval and the text local similarity between each word may be defined as follows:
text interval: for the word wi，wjIf the number of words in the interval is a, the text interval tDis (w) isi，wj)＝1+a。
In this embodiment, the text interval may be used to measure the interval distance between words in the text. When two words have a plurality of text interval values, the minimum value is taken as the text interval. Thus, when a word appears many times in the text, the word will be less text-spaced from other words, and if the first and last words appear only a few times in the text, it will be much more text-spaced from other words.
Text local similarity: according to the characteristics of the keywords: words or phrases in text with importance and theme, defining word wi，wjLocal similarity of text under the current text
Wherein, sim (w)i，wj) The similarity between two word vectors can be calculated by cosine distance or Euclidean distance. Specifying the word wiThe local similarity of the text to itself is 0.
In this embodiment, the local similarity of the text may be used to measure the similarity between words in the text, and when the similarity between two words is determined, the larger the text interval is, the smaller the local similarity is, and otherwise, the larger the local similarity is.
S302: and constructing an uncertainty map according to the words in the response text and the local similarity of the texts between any two words.
In this embodiment, an uncertain graph model of a text segment (short text) in a response text is constructed by calculating local similarity of texts between words.
Specifically, the uncertainty map may be represented as a triple: g ═ (V, E, p), where V ═ V1，v2，...，vnIs the set of all vertices, E ═ E1，e2，...，emIs the set of all edges, p ═ p1，p2，...，pmDenotes the probability of each edge existing. In order to simplify the model calculation, the probability of existence of each edge is assumed to be independent.
For example, fig. 6 is a schematic structural diagram of a 4-step uncertainty map provided in the embodiment of the present invention, as shown in fig. 6, the 4-step uncertainty map has 4 vertices ①, ②, ② 0, and ② 1, and has 4 edges, which are respectively a connecting edge of vertex ① and ②, a connecting edge of vertex ① and vertex ③, a connecting edge of vertex ② and vertex ④, and a connecting edge of vertex ③ and vertex ④.
Optionally, constructing the uncertainty map according to the words in the response text and the local similarity of the text between any two words may include:
constructing an uncertain graph G ═ V, E, p, wherein V is a set of all vertexes, E is a set of all edges, and p is the probability of existence of each edge
A set of (a); taking each word in the response text as a vertex; adding connecting edges between all the vertexes, namely E is V multiplied by V; the word w represented by two vertexesi，wjLocal similarity LocalSim (w) of text therebetweeni，wj) Probability of existence as a continuous edge of two vertices
Namely, it is
In this embodiment, according to the uncertain graph construction process, the words in the response text are represented by vertices, and each word in the ordered word group W in the response text is taken as a vertex, that is, V ═ W1，w2，...，wn}. Local similarity LocalSim (w) of texts between words is represented by probability of edge existence (edge probability for short)i，wj)。
S303: and extracting the key words from the response text according to the uncertain graph.
In this embodiment, after extracting keywords from each portion of the response text or the text structure based on the response text, and establishing an uncertain graph model by using features such as word vectors and semantic relationships between words, the keywords may be extracted from the response text by the following several implementation manners:
the first implementation mode comprises the following steps: and calculating the vertex density of each vertex according to the uncertain graph, and using the vertex density as a keyword evaluation index to evaluate the extraction quality of the keywords in the uncertain graph.
Specifically, extracting the keywords from the response text according to the uncertainty map may include:
determining adjacency matrix (A) of uncertain graph G ═ V, E, pq，(A)qElement (1) of
Is represented by a vertex wiTo the vertex wjThe number of paths of length q; according to the adjacency matrix (A)qDetermining the vertex density of each vertex, wherein the vertex density of each vertex represents the probability sum of other vertices transferred to the vertex; and obtaining a set of the keywords of the response text according to the vertex density.
In this embodiment, for a definite undirected graph, assume that its adjacency matrix is (A)qThen (A)qElement (1) of
Is represented by a vertex wiTo the vertex wjThe number of paths of length q. Can also be understood as being defined by the vertex wiStarting to go through q step lengths to a vertex wjThe number of policies of (2). FIG. 6 shows a 4-step uncertainty map with a neighboring matrix
Wherein A is understood to be (A)qStep q in (1).
Alternatively, an adjacency matrix (a) of the uncertain graph G ═ V, E, p is determinedqBefore, may also include: and removing edges with the edge probability less than or equal to 0 in the uncertain graph G (V, E, p), and deleting vertexes without connected edges after the removal is finished.
In this embodiment, the edge with the edge probability less than or equal to 0 may be pruned, and after pruning is completed, the vertex without the edge may be deleted, so as to obtain the final text uncertainty map DG. If the vertex V' of DG at this time is { w }1，w2，...，wmAn adjacency matrix a with a step q of 1 can be represented as:
in this embodiment, the definition of the correlation between the vertex of the cluster center in the deterministic graph cluster is extended to the indeterminate graph DG. For a vertex, the vertex density may be considered higher when the sum of the probabilities of other vertices transferring to the vertex is greater. Wherein the vertex (per word) w is given a step size qiVertex density ofCan be defined as:
in particular, according to the adjacency matrix (A)qDetermining the vertex density of each vertex may include the following two implementations:
implementation mode 1: and (3) a candidate keyword extraction algorithm under a certain step length, wherein the algorithm calculates the vertex density of the corresponding vertex of each word under the certain step length.
Specifically, calculating the vertex density of each vertex by a certain step size may include: using a formula
Calculating the vertex density (w) of each vertex under the preset step length qi)，r＝1，2...q。
Accordingly, deriving a set of answer text keywords from the vertex densities may include: normalizing the vertex density of all the vertexes to obtain a set S { (w) of the response text keywords1，DEN(w1))，(w2，DEN(w2))，...，(wm，DEN(wm))}，DEN(wi) Representing each word (i.e. vertex) wiNormalized vertex density of (a).
In this embodiment, at a certain step length, wiThe peak density of (d) is called DEN (w) after normalization with e.g. a min-max functioni) DEN (w)i) As the evaluation index of the candidate keyword, DEN (w) of each word under a certain step length is calculatedi) A set of candidate keywords may be obtained.
In this embodiment and the following embodiments, when obtaining the set S of the response text keywords, the vertex density DEN (w) in the set S may be seti) The vertex larger than the threshold value is used as a keyword; the set S may also be sorted from large to small according to the vertex density, and a preset number of vertices arranged in front of the set may be used as the keywords.
In the embodiment, a candidate keyword extraction algorithm under a certain step length is adopted, and vertex density is calculated based on the uncertain graph to obtain the weight of the keywords in the uncertain graph so as to evaluate the extraction quality of the keywords in the uncertain graph.
Implementation mode 2: and the self-adaptive candidate keyword extraction algorithm obtains the vertex density of the corresponding vertex of each word through step self-adaptive automatic calculation.
Specifically, the step size adaptive iteration calculation of the vertex density of each vertex through the vertex may include: :
presetting an iteration upper limit and an iteration step length iter, wherein the initial value of iter is 1; when the iteration number is less than or equal to the iteration upper limit, adopting a formula
Calculating the vertex density (w) of each vertex at the current step size iteri) Sequencing the vertex densities of all the vertexes according to a preset sequence, wherein r is 1, 2.. iter; when the vertex density ordering of the current iteration is the same as the vertex density ordering of the last iteration, exiting the iteration loop; otherwise, adding 1 to the iteration step size iter, and continuing the iteration.
Accordingly, according to the vertex density DEN (w)i) Obtaining a set of answer text keywords may include: normalizing the vertex densities of all the vertexes to obtain a set S { (w) of the response text keywords1，DEN(w1))，(w2，DEN(w2))，...，(wm，DEN(wm))}；DEN(wi) For each word wiNormalized vertex density of (a).
In this embodiment, the algorithm initialization step size may be set to 1, the current vertex density ordering is calculated in each loop process, the step sizes are sequentially increased, when the vertex density ordering order does not change in a certain loop, or when the iteration number reaches the upper limit, the loop exits, and the words corresponding to all vertices and the DEN (w) thereof are output after all vertex densities are normalizedi) The value is obtained.
In the embodiment, a self-adaptive candidate keyword extraction algorithm is adopted, and iterative computation of vertex density is performed based on the uncertain graph to obtain the weight of the keywords in the uncertain graph so as to evaluate the extraction quality of the keywords in the uncertain graph.
The second implementation mode comprises the following steps: calculating vertex density of each vertex according to the uncertainty map, and determining frequency value IDF (w) of each word in the response text based on TF-IDFi) And determining a score TRank (w) of each word in the answer text based on the TextRank algorithmi) The vertex density and frequency value IDF (w)i) And score value TRank (w)i) And combining the evaluation indexes as keywords to evaluate the extraction quality of the keywords in the uncertain graphs.
Specifically, the auxiliary response method for the intelligent customer service provided by the embodiment of the present invention may further include:
determining a frequency value IDF (w) for each word in the response text based on the TF-IDFi) And determining a score TRank (w) of each word in the answer text based on the TextRank algorithmi) (ii) a The frequency value IDF (w) of each wordi) And score value TRank (w)i) Respectively carrying out normalization to obtain normalized frequency values IDF*(wi) And normalized score TRank*(wi)。
Accordingly, deriving a set of answer text keywords from the vertex densities may include:
normalizing the vertex density of each vertex to obtain a normalized vertex density DEN (w)i) (ii) a Using the formula WDEN (w)i)＝α·IDF*(wi)+β·TextRank*(wi)+(1-α-β)DEN(wi) Determining a multi-eigenvalue WDEN (w) for each wordi) (ii) a Get the set of answer text keywords S { (w)1，WDEN(w1))，(w2，WDEN(w2))，…，(wm，WDEN(wm))}。
The values of the weights α and β may be obtained through an experiment of training set data, and are not limited and described herein.
In the prior art, some common words are ranked at the top, namely, words with a large IDF value, in the candidate keyword ranking obtained after the adaptive candidate keyword extraction algorithm processing. Therefore, the invention further carries out evaluation indexes on the candidate keywordsOptimizing, comprehensively considering influence of word frequency, word-to-word relation and co-occurrence frequency of words on keyword extraction, and enabling the vertex density DEN (w)i) Frequency value IDF (w)i) And score value TRank (w)i) A plurality of features are combined, wherein IDF (w)i) The value is used to reflect word ordering, TRank (w)i) The value is used to reflect the co-occurrence frequency of the words DEN (w)i) The value is used for reflecting the inter-word semantics, the three are integrated to obtain the keyword index with weight, and the extraction performance of the response text keyword is better.
Specifically, for the ordered phrase W of the response text, { W ═ W1，w2，...，wmFirst, the IDF value is normalized by the min-max function to define the IDF (w)i) Is wiNormalized IDF value. And its TextRank value are also normalized by the min-max function, defining TRank (w)i) Is wiNormalized TextRank values. Wherein the vertex density DEN (w)i) The calculation can be described in detail in the above embodiments, which are not described herein again. Combining the three to obtain a weighted keyword index WDEN (w)i)＝α·IDF*(wi)+β·TextRank*(wi)+(1-α-β)DEN(wi)。
In this embodiment, when obtaining the set S of the response text keywords, the multi-eigenvalue WDEN (w) in the set S may be obtainedi) The vertex larger than the threshold value is used as a keyword; the set S may also be sorted from large to small according to the multi-eigenvalue, and a preset number of vertices arranged in front of the set may be used as the keywords.
In this embodiment, the vertex density DEN (w)i) Frequency value IDF (w)i) And score value TRank (w)i) And a plurality of characteristics are combined to extract the keywords with weights, so that the extraction performance of the keywords of the response text is better.
The auxiliary response method of the intelligent customer service provided by the invention extracts keywords from each part of the response text or the response text, establishes an uncertain graph model by combining the characteristics of word vectors, the semantic relation among words, the text structure of the application text and the like, combines the word frequency, the co-occurrence frequency among words and the vertex density in the uncertain graph model with the characteristics of optimizing the keyword extraction index, obviously improves the performance of keyword extraction, and solves the problem of low performance of keyword extraction in the prior art.
The invention also provides an auxiliary response system of the intelligent customer service, which is used for assisting the customer service staff to make a response. Fig. 7 is a schematic structural diagram of an auxiliary response system for intelligent customer service provided in an embodiment of the present invention, and as shown in fig. 7, the auxiliary response system for intelligent customer service provided in an embodiment of the present invention may include: a decimation module 71, a matching module 72 and a feedback module 73.
The extraction module 71 is used for acquiring the user consultation information and extracting the keywords from the response text based on the interword relationship;
a matching module 72, configured to locate, based on the keyword, response knowledge in the response text that matches the user consultation information;
and the feedback module 73 is used for feeding the response knowledge back to the customer service staff so that the customer service staff can respond to the user consultation information according to the response knowledge.
The auxiliary response system of the intelligent customer service provided by the embodiment of the present invention is used for executing the technical solution of the method embodiment shown in fig. 1, and the implementation principle and the implementation effect thereof are similar, and are not described herein again.
Further, in the above-described embodiment, the inter-word relationship represents a local similarity of text between words;
the extraction module 71 extracts keywords from the response text based on the inter-word relationship, and may include:
performing word segmentation and word vectorization on the response text, and determining the local similarity of the text between any two words in the response text;
constructing an uncertain graph according to words in the response text and the text local similarity between any two words;
and extracting the key words from the response text according to the uncertain graph.
Further, in the above embodiment, the extracting module 71 constructs the uncertainty map according to the words in the answer text and the local similarity of the text between any two words, and may include:
constructing an uncertain graph G ═ V, E, p, wherein V is a set of all vertexes, E is a set of all edges, and p is the probability of existence of each edge
A set of (a);
taking each word in the response text as a vertex;
adding connecting edges between all the vertexes, namely E is V multiplied by V;
the word w represented by two vertexesi，wjLocal similarity LocalSim (w) of text therebetweeni，wj) Probability of existence as a continuous edge of two vertices
Namely, it is
Further, in the above embodiment, the extracting module 71 extracts the keywords from the response text according to the uncertainty map, and may include:
determining adjacency matrix (A) of uncertain graph G ═ V, E, pq，(A)qElement (1) of
Is represented by a vertex wiTo the vertex wjThe number of paths of length q;
according to the adjacency matrix (A)qDetermining the vertex density of each vertex, wherein the vertex density of each vertex represents the probability sum of other vertices transferred to the vertex;
and obtaining a set of the keywords of the response text according to the vertex density.
Further, in the above embodiment, the decimation module 71 is based on the adjacency matrix (A)qDetermining the vertex density for each vertex may include:
using a formula
Calculating the vertex density dense (w) of each vertex under the preset step length qi)，r＝1，2...q；
Accordingly, the extracting module 71 obtains a set of keywords of the response text according to the vertex density, which may include:
normalizing the vertex density of all the vertexes to obtain a set S { (w) of the response text keywords1，DEN(w1))，(w2，DEN(w2))，...，(wm，DEN(wm))}，DEN(wi) Represents each word wiNormalized vertex density of (a).
Further, in the above embodiment, the decimation module 71 is based on the adjacency matrix (A)qDetermining the vertex density for each vertex may include:
and (3) calculating the vertex density of each vertex through the self-adaptive iteration of the step length between the vertices:
presetting an iteration upper limit and an iteration step length iter, wherein the initial value of iter is 1;
when the iteration number is less than or equal to the iteration upper limit, adopting a formula
Calculating the vertex density (w) of each vertex at the current step size iteri) Sequencing the vertex densities of all the vertexes according to a preset sequence, wherein r is 1, 2.. iter;
when the vertex density ordering of the current iteration is the same as the vertex density ordering of the last iteration, exiting the iteration loop; otherwise, adding 1 to the iteration step size iter, and continuing the iteration.
Accordingly, the decimation module 71 depends on the vertex density DEN (w)i) Obtaining a set of answer text keywords may include:
normalizing the vertex densities of all the vertexes to obtain a set S = { (w) of the response text keywords1，DEN(w1))，(w2，DEN(w2))，...，(wm，DEN(wm))}；DEN(wi) For each word wiNormalized vertex density of (a).
Further, in the foregoing embodiment, the auxiliary response system for intelligent customer service provided in the embodiment of the present invention may further include:
an IDF module for determining a frequency value IDF (w) for each word in the response text based on the TF-IDFi) Frequency value IDF (w) of each wordi) Normalizing to obtain normalized frequency value IDF*(wi)；
A TRank module for determining the score TRank (w) of each word in the response text based on the TextRank algorithmi) The score TRank (w) of each wordi) Normalization is carried out to obtain a normalized score TRank*(wi)；
Accordingly, the extracting module 71 obtains a set of keywords of the response text according to the vertex density, which may include:
normalizing the vertex density of each vertex to obtain a normalized vertex density DEN (w)i)；
Using the formula WDEN (w)i)＝α·IDF*(wi)+β·TextRank*(wi)+(1-α-β)DEN(wi) Determining a multi-eigenvalue WDEN (w) for each wordi) A set of answer text keywords S { (w) is obtained1，WDEN(w1))，(w2，WDEN(w2))，...，(wm，WDEN(wm))}。
Further, in the foregoing embodiment, the extracting module 71 performs word segmentation and word vectorization on the response text, and determines a local similarity of the text between any two words in the response text, which may include:
performing word segmentation and stop word removal on the response text to obtain an ordered word group W with the length of n ═ W1，w2，w3，...，wn}；
Determining two words w based on a word vectoriAnd wjSimilarity between vectors sim (w)i，wj)；
Determining two words wiAnd wjText interval tDis (w) in betweeni，wj)；
According to sim (w)i，wj) And tDis (w)i，wj) Determination of wiAnd wjLocal similarity LocalSim (w) of text therebetweeni，wj)；
Wherein,
tDis(wi，wj) 1+ a, a denotes two words wiAnd wjThe number of words in the interval between.
Further, in the above-described embodiment, the response text includes a product specification, a product operation manual, a product order information table, or an after-market service manual.
It will be understood by those of ordinary skill in the art that all or some of the steps of the methods, systems, functional modules/units in the devices disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof. In a hardware implementation, the division between functional modules/units mentioned in the above description does not necessarily correspond to the division of physical components; for example, one physical component may have multiple functions, or one function or step may be performed by several physical components in cooperation. Some or all of the components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). The term computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as is well known to those of ordinary skill in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, Digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can accessed by a computer. In addition, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media as known to those skilled in the art.
Claims (10)
1. An auxiliary response method of intelligent customer service, which is used for assisting the customer service personnel to make a response, and is characterized in that the method comprises the following steps:
acquiring user consultation information, and extracting keywords from the response text based on the interword relationship;
positioning response knowledge matched with the user consultation information in the response text based on the keyword;
and feeding back the response knowledge to customer service staff so that the customer service staff can respond to the user consultation information according to the response knowledge.
2. The method of claim 1, wherein the inter-word relationships represent local similarities of text between words;
the extraction of the keywords from the response text based on the word relation comprises the following steps:
performing word segmentation and word vectorization on the response text, and determining the local similarity of the text between any two words in the response text;
constructing an uncertain graph according to words in the response text and the text local similarity between any two words; (ii) a
And extracting the key words from the response text according to the uncertain graph.
3. The method of claim 2, wherein constructing the uncertainty map based on local similarity of text between a word and any two words in the response text comprises:
constructing uncertainty map G ═ V, E, p, V is the set of all vertices, E is the vertexSet of edges, p being the probability of each edge being present
A set of (a);
taking each word in the response text as a vertex;
adding connecting edges between all the vertexes, namely E is V multiplied by V;
the word w represented by two vertexesi，wjLocal similarity LocalSim (w) of text therebetweeni，wj) Probability of existence as a continuous edge of two vertices
Namely, it is
4. The method of claim 3, wherein extracting keywords from the response text based on the uncertainty map comprises:
determining adjacency matrix (A) of uncertain graph G ═ V, E, pq，(A)qElement (1) of
Is represented by a vertex wiTo the vertex wjThe number of paths of length q;
according to the adjacency matrix (A)qDetermining the vertex density of each vertex, wherein the vertex density of each vertex represents the probability sum of other vertices transferred to the vertex;
and obtaining a set of the keywords of the response text according to the vertex density.
5. Method according to claim 4, characterized in that said method is based on an adjacency matrix (A)qDetermining a vertex density for each vertex, comprising:
using a formula
Calculating the vertex density (w) of each vertex under the preset step length qi)，r＝1，2...q；
The obtaining of the set of the response text keywords according to the vertex density includes:
normalizing the vertex density of all the vertexes to obtain a set S { (w) of the response text keywords1，DEN(w1))，(w2，DEN(w2))，...，(wm，DEN(wm))}，DEN(wi) Represents each word wiNormalized vertex density of (a).
6. Method according to claim 4, characterized in that said method is based on an adjacency matrix (A)qDetermining a vertex density for each vertex, comprising:
and (3) calculating the vertex density of each vertex through the self-adaptive iteration of the step length between the vertices:
presetting an iteration upper limit and an iteration step length iter, wherein the initial value of iter is 1;
when the iteration number is less than or equal to the iteration upper limit, adopting a formula
Calculating the vertex density (w) of each vertex at the current step size iteri) Sequencing the vertex densities of all the vertexes according to a preset sequence, wherein r is 1, 2.. iter;
when the vertex density ordering of the current iteration is the same as the vertex density ordering of the last iteration, exiting the iteration loop; otherwise, adding 1 to the iteration step size iter, and continuing iteration;
the DEN (w) according to the vertex densityi) Obtaining a set of answer text keywords comprising:
normalizing the vertex densities of all the vertexes to obtain a set S { (w) of the response text keywords1，DEN(w1))，(w2，DEN(w2))，...，(wm，DEN(wm))}；DEN(wi) For each word wiNormalized vertex density of (a).
7. The method of claim 4, further comprising:
determining a frequency value IDF (w) of each word in the answer text based on a word frequency-inverse document frequency algorithm TF-IDFi) And determining a score TRank (w) of each word in the answer text based on the TextRank algorithmi)；
The frequency value IDF (w) of each wordi) And score value TRank (w)i) Respectively carrying out normalization to obtain normalized frequency values IDF*(wi) And normalized score TRank*(wi)；
The obtaining of the set of the response text keywords according to the vertex density includes:
normalizing the vertex density of each vertex to obtain a normalized vertex density DEN (w)i)；
Using the formula WDEN (w)i)＝α·IDF*(wi)+β·TextRank*(wi)+(1-α-β)DEN(wi) Determining a multi-eigenvalue WDEN (w) for each wordi) A set of answer text keywords S { (w) is obtained1，WDEN(w1))，(w２，WDEN(w２))，(w1，WDEN(w2))，...，(wm，WDEN(wm))}。
8. The method according to any one of claims 4-7, wherein the performing word segmentation and word vectorization on the response text and determining local similarity of text between any two words in the response text comprises:
performing word segmentation and stop word removal on the response text to obtain an ordered word group W with the length of n ═ W1，w2，w3，...，wn}；
Determining two words w based on a word vectoriAnd wjVector quantitySimilarity between them sim (w)i，wj)；
Determining two words wiAnd wjText interval tDis (w) in betweeni，wj)；
According to sim (w)i，wj) And tDis (w)i，wj) Determination of wiAnd wjLocal similarity LocalSim (w) of text therebetweeni，wj)；
Wherein,
tDis(wi，wj) 1+ a, a denotes two words wiAnd wjThe number of words in the interval between.
9. The method of claim 1, wherein the response text comprises a product specification, a product operation manual, a product order information sheet, or an after-market service manual.
10. An auxiliary response system of intelligent customer service, which is used for assisting the customer service personnel to make a response, and is characterized by comprising:
the extraction module is used for acquiring the user consultation information and extracting the key words from the response text based on the inter-word relation;
the matching module is used for positioning response knowledge matched with the user consultation information in the response text based on the keywords;
and the feedback module is used for feeding the response knowledge back to customer service staff so that the customer service staff can respond to the user consultation information according to the response knowledge.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911087804.9A CN111078832A (en) | 2019-11-08 | 2019-11-08 | Auxiliary response method and system for intelligent customer service |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911087804.9A CN111078832A (en) | 2019-11-08 | 2019-11-08 | Auxiliary response method and system for intelligent customer service |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111078832A true CN111078832A (en) | 2020-04-28 |
Family
ID=70310733
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911087804.9A Pending CN111078832A (en) | 2019-11-08 | 2019-11-08 | Auxiliary response method and system for intelligent customer service |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111078832A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112596646A (en) * | 2020-12-21 | 2021-04-02 | 维沃移动通信有限公司 | Information display method and device and electronic equipment |
| CN112784030A (en) * | 2021-01-28 | 2021-05-11 | 北京三快在线科技有限公司 | Method and device for generating sample, storage medium and electronic equipment |
| CN113191145A (en) * | 2021-05-21 | 2021-07-30 | 百度在线网络技术（北京）有限公司 | Keyword processing method and device, electronic equipment and medium |
| CN115907785A (en) * | 2022-11-22 | 2023-04-04 | 深圳市米糠云科技有限公司 | Intelligent customer service question and answer assisting method and device based on associated keywords |
| CN116302565A (en) * | 2023-05-06 | 2023-06-23 | 北京国联政信科技有限公司 | Digital information transmission method, storage medium and electronic equipment |
| CN117151122A (en) * | 2023-10-30 | 2023-12-01 | 湖南三湘银行股份有限公司 | Bank customer service session question-answering processing method and system based on natural language processing |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6278996B1 (en) * | 1997-03-31 | 2001-08-21 | Brightware, Inc. | System and method for message process and response |
| CN107807960A (en) * | 2017-09-30 | 2018-03-16 | 平安科技（深圳）有限公司 | Intelligent customer service method, electronic installation and computer-readable recording medium |
| CN110162611A (en) * | 2019-04-23 | 2019-08-23 | 苏宁易购集团股份有限公司 | A kind of intelligent customer service answer method and system |
-
2019
- 2019-11-08 CN CN201911087804.9A patent/CN111078832A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6278996B1 (en) * | 1997-03-31 | 2001-08-21 | Brightware, Inc. | System and method for message process and response |
| CN107807960A (en) * | 2017-09-30 | 2018-03-16 | 平安科技（深圳）有限公司 | Intelligent customer service method, electronic installation and computer-readable recording medium |
| CN110162611A (en) * | 2019-04-23 | 2019-08-23 | 苏宁易购集团股份有限公司 | A kind of intelligent customer service answer method and system |
Non-Patent Citations (1)
| Title |
|---|
| 黄睿智等: "词间关系的不确定图模型与关键词自动抽取方法", 《小型微型计算机系统》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112596646A (en) * | 2020-12-21 | 2021-04-02 | 维沃移动通信有限公司 | Information display method and device and electronic equipment |
| CN112596646B (en) * | 2020-12-21 | 2022-05-20 | 维沃移动通信有限公司 | Information display method and device and electronic equipment |
| CN112784030A (en) * | 2021-01-28 | 2021-05-11 | 北京三快在线科技有限公司 | Method and device for generating sample, storage medium and electronic equipment |
| CN112784030B (en) * | 2021-01-28 | 2022-08-26 | 北京三快在线科技有限公司 | Method and device for generating sample, storage medium and electronic equipment |
| CN113191145A (en) * | 2021-05-21 | 2021-07-30 | 百度在线网络技术（北京）有限公司 | Keyword processing method and device, electronic equipment and medium |
| CN113191145B (en) * | 2021-05-21 | 2023-08-11 | 百度在线网络技术（北京）有限公司 | Keyword processing method and device, electronic equipment and medium |
| CN115907785A (en) * | 2022-11-22 | 2023-04-04 | 深圳市米糠云科技有限公司 | Intelligent customer service question and answer assisting method and device based on associated keywords |
| CN115907785B (en) * | 2022-11-22 | 2023-09-19 | 深圳市米糠云科技有限公司 | Customer service question-answering intelligent auxiliary method and device based on associated keywords |
| CN116302565A (en) * | 2023-05-06 | 2023-06-23 | 北京国联政信科技有限公司 | Digital information transmission method, storage medium and electronic equipment |
| CN116302565B (en) * | 2023-05-06 | 2023-08-08 | 北京国联政信科技有限公司 | Digital information transmission method, storage medium and electronic equipment |
| CN117151122A (en) * | 2023-10-30 | 2023-12-01 | 湖南三湘银行股份有限公司 | Bank customer service session question-answering processing method and system based on natural language processing |
| CN117151122B (en) * | 2023-10-30 | 2024-03-22 | 湖南三湘银行股份有限公司 | Bank customer service session question-answering processing method and system based on natural language processing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109992646B (en) | Text label extraction method and device | |
| CN111078832A (en) | Auxiliary response method and system for intelligent customer service | |
| CN109635273B (en) | Text keyword extraction method, device, equipment and storage medium | |
| CN110019732B (en) | Intelligent question answering method and related device | |
| WO2018157805A1 (en) | Automatic questioning and answering processing method and automatic questioning and answering system | |
| CN109815487B (en) | Text quality inspection method, electronic device, computer equipment and storage medium | |
| CN111159359B (en) | Document retrieval method, device and computer readable storage medium | |
| US20210117625A1 (en) | Semantic parsing of natural language query | |
| CN107862027A (en) | Retrieve intension recognizing method, device, electronic equipment and readable storage medium storing program for executing | |
| CN104199965B (en) | Semantic information retrieval method | |
| CN111191442B (en) | Similar problem generation method, device, equipment and medium | |
| CN109902156B (en) | Entity retrieval method, storage medium and electronic device | |
| CN110347908B (en) | Voice shopping method, device, medium and electronic equipment | |
| CN110990533B (en) | Method and device for determining standard text corresponding to query text | |
| CN109829045A (en) | A kind of answering method and device | |
| CN110990532A (en) | Method and device for processing text | |
| CN113792594B (en) | Method and device for locating language fragments in video based on contrast learning | |
| CN117076636A (en) | Information query method, system and equipment for intelligent customer service | |
| CN115600593A (en) | Method and device for acquiring key content of literature | |
| CN118468061B (en) | Automatic algorithm matching and parameter optimizing method and system | |
| CN114202443A (en) | Policy classification method, device, equipment and storage medium | |
| CN111563361B (en) | Text label extraction method and device and storage medium | |
| CN111966869A (en) | Phrase extraction method and device, electronic equipment and storage medium | |
| US20090234836A1 (en) | Multi-term search result with unsupervised query segmentation method and apparatus | |
| CN116610810A (en) | Intelligent searching method and system based on regulation and control of cloud knowledge graph blood relationship |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200428 |