CN111339284A - Product intelligent matching method, device, device and readable storage medium - Google Patents

Abstract

The invention relates to the technical field of big data, and discloses an intelligent product matching method, which comprises the following steps: outputting a preset question set to a user, acquiring user response information of the user to each question in the preset question set, and performing emotion recognition on the user response information to obtain the emotion information of the user; and identifying the intention of the user response information to obtain the user intention information, determining the user purchasing intention based on the user emotion information and the user intention information, inquiring a preset product adaptation table based on the user information if the user purchasing intention is a forward intention, and outputting a product matched with the user information. The invention also discloses an intelligent product matching device, equipment and a computer readable storage medium. The invention provides more efficient and rapid intelligent product matching service for the user and improves the accuracy of product recommendation.

Description

Product intelligent matching method, device, device and readable storage medium

technical field

The present invention relates to the technical field of artificial intelligence, and in particular, to a method, apparatus, device and computer-readable storage medium for intelligent product matching.

Background technique

The number and types of insurance products in several channels mainly used by so-called smart products currently on the market are very limited. At the same time, the mainstream intelligent configuration platform can only do simple information collection and a small amount of product matching. If the amount of user information increases, the configuration will increase geometrically, and the operation cannot be substantially deployed.

User information collection is preset information for users to choose, which limits the dimension and precision of user information acquisition. In the process of user information collection, the user's emotional information and the user's true intention cannot be accurately judged, and it is easy to be misled by untrue information, so that the user's real needs cannot be determined, and the matching degree between the product and the user is not accurate enough.

SUMMARY OF THE INVENTION

The main purpose of the present invention is to provide a product intelligent matching method, device, equipment and computer-readable storage medium, which aims to solve the technical problem of low accuracy of existing user intention and user emotion recognition and improve the matching accuracy between products and users .

In order to achieve the above object, the present invention provides a product intelligent matching method, and the product intelligent matching method includes the following steps:

outputting a preset question set to the user, and obtaining the user's response information for each question in the preset question set;

Perform emotion recognition on the user's reply information to obtain the user's emotion information;

Perform intention identification on the user's reply information to obtain the user's intention information, and determine the user's purchase intention based on the user's emotion information and the user's intention information;

If the user's purchase intention is a positive intention, based on the user information, a preset product adaptation table is queried, and a product matching the user information is output.

Optionally, before the outputting the preset question set to the user, and acquiring the user response information of the user to each question in the preset question set, the method further includes:

Collect all product document information in the preset format displayed on the specified website through web crawler technology;

Performing entity relationship extraction processing on the product document information to obtain named entity objects of each product knowledge point in the product document information and entity relationships between the named entity objects;

According to the named entity object of each product knowledge point, using a regular expression to identify the entity attribute corresponding to each named entity object in the product document information;

Using the named entity objects, entity attributes and entity relationships corresponding to each product knowledge point as source data, build a product knowledge graph based on graph database;

Based on the product knowledge graph, the question set and the product adaptation table are created.

Optionally, the outputting the preset question set to the user, and acquiring the user response information of the user to each question in the preset question set includes:

Obtain the user's reply information for each round of dialogue;

matching the user reply information with the preset slots in the preset question set;

If the user reply information is successfully matched with the preset slot, extract the preset slot;

If there is answer information matching the preset slot in the user answer information, associating the answer information with the preset slot;

Based on the association result, user information is collected.

Optionally, performing emotion recognition on the user reply information to obtain the user emotion information includes:

dividing the user reply information into a plurality of data segments;

Perform emotional judgments on the data sections respectively to obtain the degree attribute values of various emotions;

Based on the emotional degree attribute value of each data section, determine the emotional label of the data section respectively;

Based on the emotion labels of the data sections, outputting the emotion label probability distribution map corresponding to each emotion;

The user emotion information is obtained based on the emotion tag probability distribution map.

Optionally, performing intention identification on the user reply information to obtain the user intention information, and determining the user purchase intention based on the user emotion information and the user intention information includes:

Analyzing the user reply information in word units to obtain a named entity as a user intent parameter candidate;

Analyzing the user reply information, and fuzzy matching word by word according to a preset user intent key candidate set to obtain intent keywords;

outputting a user intent identification result based on the intent keyword;

The purchase intention of the user is determined based on the user intention recognition result and the user emotion information.

Optionally, if the purchase intention of the user is a positive intention, query a preset product adaptation table based on the user information, and output products matching the user information include:

Translate the user information into a label format, and generate a plurality of labels corresponding to the user information;

If the purchase intention of the user is a positive intention, then based on the preset weight value of the tag, each tag is matched with the product in the preset product adaptation table, and the relationship between each tag and the corresponding product is calculated and output. match score;

Based on the matching score of each label and the corresponding product, summing the matching score of each label and the corresponding product to obtain the matching degree score of each product;

If the matching degree score of the product is higher than the preset threshold, mark the product as recommended, if not, mark it as non-recommended;

The preset number of products marked as recommended are used as recommended results and output.

Further, the present invention also provides a product intelligent matching device, the product intelligent matching device includes:

an acquisition module, outputting a preset question set to the user, and acquiring the user's response information for each question in the preset question set;

an emotion recognition module, configured to perform emotion recognition on the user response information to obtain current user emotion information;

an intention recognition module, which performs intention recognition on the user's reply information to obtain the user's intention information, and determines the user's purchase intention based on the user's emotion information and the user's intention information;

The product matching module is used for querying a preset product adaptation table based on the user information when the user's intention is a positive intention, and outputting a product matching the user information.

Optionally, the product intelligent matching device further includes:

The collection module is used to collect all the product document information in the preset format displayed on the designated website through the web crawler technology;

The building module is used to perform entity relationship extraction processing on the product document information, and obtain the named entity objects of each product knowledge point in the product document information and the entity relationship between the named entity objects; according to the named entity objects of each product knowledge point object, using regular expressions to identify the entity attributes corresponding to each named entity object in the product document information; using the named entity objects, entity attributes and entity relationships corresponding to each product knowledge point as source data to build a graph database-based product knowledge map ;

A creation module is configured to create the question set and the product adaptation table based on the product knowledge graph.

Optionally, the obtaining module is specifically used for:

Obtain the user reply information of each round of dialogue, and match the user reply information with the preset slot in the preset question set; when the user reply information is successfully matched with the preset slot, extract the Preset slot; when there is answer information matching the preset slot in the user reply information, then associate the answer information with the preset slot, and collect user information based on the association result .

Optionally, the emotion recognition module is specifically used for:

The user reply information is divided into a plurality of data sections, and emotional judgments are performed on the data sections to obtain the degree attribute values of various emotions; based on the emotional degree attribute values of the various data sections, determine the The emotion label of the data section is output, and the probability distribution map of the emotion label corresponding to each emotion is output to obtain the user emotion.

Optionally, the intention recognition module is specifically used for:

Analyzing the user reply information in units of words to obtain a named entity as a user intent parameter candidate, analyzing the user reply information, and word-by-word fuzzy matching according to a preset user intent key candidate set to obtain intent keywords, A user intent identification result is output, and the user's purchase intent is determined based on the user intent identification result and the user emotion.

Optionally, the product recommendation module is specifically used for:

Translate the user information into a tag format, and generate multiple tags corresponding to the user information. When the user's purchase intention is a positive intention, then based on the preset weight value of the tag, the tags are respectively Match with the products in the preset product adaptation table, calculate and output the matching scores of each label and the corresponding product, and sum up the matching scores of each label and the corresponding product to obtain the matching degree score of each product. When the matching degree score of the product is higher than the preset threshold, the product is marked as recommended, if not, it is marked as non-recommended; a preset number of products marked as recommended are used as recommendation results and output.

Further, in order to achieve the above object, the present invention also provides a product intelligent matching device, the product intelligent matching device includes a memory, a processor, and a product intelligent matching device stored on the memory and running on the processor. A program, when the intelligent product matching program is executed by the processor, implements the steps of any one of the above-mentioned methods for intelligent product matching.

Further, in order to achieve the above object, the present invention also provides a computer-readable storage medium on which a product intelligent matching program is stored, and when the product intelligent matching program is executed by a processor, any of the above-mentioned items are realized. The steps of a described product intelligent matching method.

The intelligent product matching method proposed by the present invention improves the current market so-called intelligent product assistant business model and technical defects, and provides a product intelligent matching method based on user information and user needs. Collect the product document information of the designated website through the web crawler technology, process the product document information, and build a problem set for product knowledge points for multiple rounds of dialogue to collect user information, so as to avoid the possibility of simple positive collection of user information. The resulting lack of information or inconsistent understanding of insurance. At the same time, it collects user personalized information based on multiple dimensions, and solves the problem of user needs and product matching through the knowledge graph question and answer of the preset model. After matching through the user & product adaptation model, accurate product recommendations can be made.

Description of drawings

1 is a schematic structural diagram of a device hardware operating environment involved in an embodiment of an intelligent product matching device solution according to the present invention;

2 is a schematic flowchart of an embodiment of an intelligent product matching method according to the present invention;

FIG. 3 is a schematic diagram of a refinement flow of an embodiment of step S110 in FIG. 2;

4 is a schematic flowchart of a second embodiment of the intelligent product matching method of the present invention;

FIG. 5 is a schematic diagram of a refinement flow of an embodiment of step S120 in FIG. 2;

FIG. 6 is a schematic diagram of a refinement flow of an embodiment of step S130 in FIG. 2;

FIG. 7 is a schematic diagram of a refinement flow of an embodiment of step S140 in FIG. 2;

FIG. 8 is a schematic diagram of functional modules of an embodiment of an intelligent product matching device according to the present invention.

The realization, functional characteristics and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments.

Detailed ways

It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

The invention provides an intelligent product matching device.

Referring to FIG. 1 , FIG. 1 is a schematic structural diagram of a device hardware operating environment involved in an embodiment of an intelligent product matching device according to the present invention.

As shown in FIG. 1 , the product intelligent matching device may include: a processor 1001 , such as a CPU, a communication bus 1002 , a user interface 1003 , a network interface 1004 , and a memory 1005 . Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. Optionally, the network interface 1004 may include a standard wired interface and a wireless interface (eg, a WI-FI interface). The memory 1005 may be high-speed RAM memory, or may be non-volatile memory, such as disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001 .

Those skilled in the art can understand that the hardware structure of the intelligent product matching device shown in FIG. Or a different component arrangement.

As shown in FIG. 1 , the memory 1005 as a computer-readable storage medium may include an operating system, a network communication module, a user interface module, and an intelligent product matching program. Among them, the operating system is a program for managing and controlling product intelligent matching equipment and software resources, and supports the operation of the network communication module, user interface module, product intelligent matching program and other programs or software; the network communication module is used to manage and control the network interface 1004 ; The user interface module is used to manage and control the user interface 1003.

In the hardware structure of the product intelligent matching device shown in FIG. 1 , the network interface 1004 is mainly used to connect to the system background and perform data communication with the system background; the user interface 1003 is mainly used to connect the client (client) and perform data communication with the client. Communication; the product intelligent matching device invokes the product intelligent matching program stored in the memory 1005 through the processor 1001, and executes the operations of the following embodiments of the product intelligent matching method.

Based on the above-mentioned hardware structure of the product intelligent matching device, various embodiments of the product intelligent matching method of the present invention are proposed.

Referring to FIG. 2 , FIG. 2 is a schematic flowchart of an embodiment of an intelligent product matching method of the present invention. In this embodiment, the intelligent product matching method includes the following steps:

Step S110, outputting the preset question set to the user, and acquiring the user response information of the user to each question in the preset question set;

In this embodiment, the question set is an automatically generated question and answer library for each knowledge point of the insurance document through the analysis and extraction of the insurance knowledge in the insurance product document through the reading comprehension model, which can support multiple rounds of dialogues to collect user information. It further avoids the lack of information or inconsistent understanding of insurance that may be caused by simply collecting user information in a forward direction. Based on the user response information collected in each round of the multi-round dialogue, the user information is collected in multiple dimensions.

In this embodiment, the user information refers to the user's personal information, family relationship information, family asset status, risk allocation information, insurance usage information, and the like. Provide users with more accurate insurance product configuration based on multi-dimensional user information.

Step S120, performing emotion recognition on the user reply information to obtain the user emotion information;

In this embodiment, the user's reply information refers to the user's reply information collected by conducting a dialogue according to a preset set of questions during multiple rounds of dialogue. Perform emotion recognition on the user's reply information, extract audio features or semantic features of the user's reply information, etc., and determine the user's emotion information according to the emotion recognition result.

In this embodiment, the user's emotional information refers to the emotional information when the user answers the question during multiple rounds of dialogue. If the user's answer is negative, the target question needs to be euphemistically confirmed twice in the multiple rounds of dialogue, so as to obtain the user's emotional response. real needs. For example, in multiple rounds of conversations, the system asks "Do you want to know about critical illness insurance?" If the user answers "Yes, please introduce the details of critical illness insurance", it is determined that the user's emotional information is positive. To understand "critical illness insurance", it also shows that users have certain purchase needs for critical illness insurance. If the user answers "Is it okay" or "I can learn about it", it is determined that the user is not interested in critical illness insurance, and the emotional information is negative.

Step S130, performing intention identification on the user reply information to obtain the user intention information, and determining the user's purchase intention based on the user emotion information and the user intention information;

In this embodiment, performing intention recognition on the user's reply information refers to performing semantic recognition and text analysis on the user's reply information, and judging the basic intention information of the user according to the results of the semantic recognition and text analysis. Then, according to the emotion recognition results, the purchase intention of the user is determined. For example, the system asks "Are you familiar with the specific terms and conditions of our company's children's insurance", and the user's reply information is "I don't know much, can you introduce the children's insurance in detail?" Semantic analysis and text analysis are performed on the user's response information to determine The user intention information is: "The user does not understand children's insurance", "The user wants to know the specific terms of the children's insurance", and it can also be judged that the user's attitude is positive. According to the user's intention information and the user's emotional information, Further determine the user's purchase intention: the user has the intention to purchase children's insurance.

Step S140, if the user's purchase intention is a positive intention, query a preset product adaptation table based on the user information, and output a product matching the user information.

In this embodiment, the positive intention can also be said to be the real purchase intention. If the user's purchase intention is positive, it means that the user is interested in the product and has purchase intention and demand. According to the collected user information, query the preset product adaptation table, and output products that match the user's purchasing needs.

In this embodiment, the product adaptation table refers to a collection of insurance products suitable for the user designed according to multi-dimensional information such as the user's age, work nature, income status (current and expected), family status, risk preference, and financial ability. For example, according to the user's information, the user can buy products with the adaptation table A, C, S, H, K, but the user's demand is to buy products A, D, H, according to the user's positive purchase intention and User information, query product adaptation table, and output matching products.

In this embodiment, multiple rounds of dialogue are conducted with the user through the question set generated by the insurance document knowledge points, and more accurate user information is collected based on multiple dimensions, which avoids preset information for the user to choose, and the user information acquisition dimension and The problem of limited accuracy. At the same time, in the process of collecting user information in multiple rounds of dialogue, the ability to use emotion recognition is increased, and the emotion carried in the user's reply information is recognized, and the user's emotion is judged. If the user's emotion is positive, it can be determined that the answer is the user's true intention; if the user's emotion is relatively negative, a euphemistic secondary confirmation is performed to confirm the user's true intention and improve the accuracy of the user's true intention judgment. So as to obtain the real needs of users and avoid being misled by untrue information.

Further, the user's intention is identified according to the user's reply information, and the user's intention identification result is obtained. Combining user intentions and user emotions, querying the product adaptation table based on user information, and outputting products that are more suitable for users. Among them, the product adaptation table is a collection of insurance products that are more suitable for users to purchase based on multi-dimensional information such as the user's age, work nature, income status (current and expected), family status, risk preference, and financial ability. , according to the subjective needs of users, configure suitable products and recommend them to users.

Referring to FIG. 3 , FIG. 3 is a schematic diagram of a refinement flow of an embodiment of step S110 in FIG. 2 . In this embodiment, the above step S110 includes:

Step S1101, obtaining user reply information for each round of dialogue;

In this embodiment, the multi-round dialogue is suitable for closed scenarios, which is a method of obtaining necessary information to finally obtain a clear user instruction after initially clarifying the user's intention in the human-machine dialogue. According to the user's reply information collected in multiple rounds of dialogue, the information contained in it itself only accounts for a small part of the total amount of information transmitted, and more information comes from a series of scene information such as the speaker's identity and current time/place. Therefore, the information acquisition method of multi-round dialogue should not be limited to what the user said.

Multi-turn dialogues do not necessarily represent multiple dialogue interactions with the user in form, for example, if sufficient information has been provided in the user's utterance, or supplementary information from other sources is sufficient to convert the user's initial intention into a sentence. With explicit user instructions, there will be no multiple conversational interactions with the user. Obtain user response information according to multiple rounds of dialogue.

In this embodiment, the user reply information may be the user's personal basic information, or may be the user's risk configuration preference information or the like.

Step S1102, matching the user reply information with a preset slot in a preset question set;

In this embodiment, the acquired user reply information is matched with the preset slots in the preset question set.

In this embodiment, the slot is the information that needs to be completed to convert the preliminary user intention into a clear user instruction in the process of multiple rounds of dialogue. A slot corresponds to a kind of information that needs to be obtained in the processing of one thing. Among them, it should be noted that not all slots must be successfully matched and associated. Take the following dialogue as an example: me: [how much does it cost to go to Xiaoshan Airport], taxi driver: [70]; [70] in the dialogue , should be understood as RMB 70, and there is no need to ask: [Are you talking about RMB, US dollar, Japanese yen or Hong Kong dollar? ]. This type of information should exist in the form of a default value, that is, the slot has a necessary association and a non-essential association, which corresponds to the above-mentioned [information does not necessarily need to be obtained through a dialogue with the user].

Step S1103, if the user reply information is successfully matched with the preset slot, extract the preset slot;

In this embodiment, the slots are divided into two slot types: word slots and interface slots. If the user's reply information matches the preset slot successfully, the successfully matched slot is extracted. For example, if the user expresses "no", A series of statements such as "No", "No", "No" can be successfully matched with the slot [No], then the slot will be extracted.

Step S1104, if there is answer information matching the preset slot in the user reply information, then associate the answer information with the preset slot, and output an association result;

In this embodiment, if there is answer information matching the successfully matched slot in the user's reply information, the relevant answer information is filled in the slot and output. There are two situations to discuss here. One is: We clearly know the relevant answer information and can directly fill in the slot without confirming to the user. For example, in multiple rounds of dialogue, the system asks the user "How is your household income?" The user answers "My annual income is about 100,000", then it will be successfully matched with the "Annual Income" slot. The other is: we know that the relevant answer information can only be used as a reference, and the user's assistance can be used to fill in the slot. For example, the user expresses a series of statements such as "no", "no", "no", "no" and so on. Both can be successfully matched with the slot [Not]. In this case, an option needs to be provided so that the user can finally decide the filling value of the slot.

Step S1105: Collect user information based on the association result.

In this embodiment, if the user's reply information is successfully matched with the preset slot in the question set, it is said that the user's reply information is associated with the slot. For example, in multiple rounds of dialogue, the system asks the user "How is your household income?" The user answers "My annual income is about 100,000", then it will be successfully matched with the "Annual Income" slot. Associate "100,000" in the user's answer with this slot, and output the associated result. The association result is "annual income of 100,000 yuan", and user information is collected according to the association result, and the user's annual income is 100,000 yuan.

Referring to FIG. 4 , FIG. 4 is a schematic flowchart of the second embodiment of the intelligent product matching method of the present invention. In this embodiment, before the above step S110, the method further includes:

Step S210, collecting all product document information in a preset format displayed on a designated website through a web crawler technology;

In this embodiment, tens of billions of web page information in the Internet can be saved locally by means of the web crawler technology. Specifically, the crawler code program simulates the browser to send a request to the network server, so as to read the network resources from the network stream and save it locally. In addition, based on the relevant information extraction rules, the user needs are extracted from the crawled information. Information.

In this embodiment, based on the different types of websites to be crawled, the content of the collected product document information is also different. For example, if data is crawled from a certain website, the content collected is property insurance related information, while if data is crawled from the official website of a technology company, the content collected is scientific and technological knowledge, such as big data management, Data actuarial, AI, etc. Based on the different crawled content, the product knowledge graphs constructed are also different.

In this embodiment, the crawling method is not limited. It is preferable to deploy a specified crawler program through a Docker container as a medium to crawl the specified website content. For example, crawling insurance product information on the website of an insurance company, the price of the corresponding product, and the corresponding purchasers. The crawled websites include designated public websites and websites searched by search engines.

In this embodiment, the latest insurance product documents are crawled from the official websites of all insurance companies in the market through the web crawler technology, and the insurance status information is added. The crawler crawls once a week to update the status of insurance products, so as to obtain the document information of insurance products on the whole network.

In this embodiment, the insurance status information refers to the sales status information of the insurance product, including the information of newly launched, on sale, and discontinued sale. On this basis, the information of the insurance sales area and the local branch of the insurance company is added. The set of insurance products recommended to users can be determined only by the sales status and sales region. Otherwise, some products have been discontinued or the region does not support them, and this type of insurance product is not suitable for recommendation no matter how good it is.

Step S220, performing entity relationship extraction processing on the product data to obtain named entity objects of each product knowledge point in the product document information and entity relationships between named entity objects;

In this embodiment, in order to facilitate the generation of a question set for each knowledge point of the insurance document according to the product knowledge points, and to support the reverse confirmation of user information in multiple rounds of dialogue, it is necessary to obtain the product knowledge points and the entity relationship between the various product knowledge points in advance. In this embodiment, the natural language processing technology is used to extract the entity relationship, that is, the named entity object of each product knowledge point in the product document information and the entity relationship between the named entity objects are extracted.

The main purpose of Natural Language Processing (NLP) technology is to help machines better understand human language, including basic lexical, syntactic and other semantic understanding, as well as high-level understanding such as needs and emotions, and then make up for human communication ( The gap between natural language) and computer understanding (machine language).

In this embodiment, after crawling the relevant content pre-specified in the website, it is necessary to extract product knowledge points from the crawled content through natural language processing technology, for example, product price information, product term information, product common words and commonly used products Questions and answers and related referenced product documentation content, etc.

In this embodiment, knowledge points such as corresponding insurance product details, insurance product proper nouns, and commonly used insurance product concepts are acquired through the knowledge extraction technology based on natural language processing. Specifically, knowledge points are extracted from the crawled website content based on the product name, status information and compensation scope in the insurance product document information.

Step S230, according to the named entity object of each product knowledge point, using a regular expression to identify the entity attribute corresponding to each named entity object in the product document information;

In this embodiment, in this embodiment, in order to construct a product knowledge structure diagram, after extracting the named entity objects of each product knowledge point in the product document information, a regular expression is further used to identify the corresponding named entity objects in the product document. Entity properties.

Regular expressions describe a pattern of string matching, which can be used to check whether a string contains a certain substring, replace a matched substring, or extract a substring that meets a certain condition from a string, etc. Its design idea is to use a descriptive language to define a rule for a string, and any string that conforms to the rule is considered to "match" the regular expression.

In this embodiment, a regular expression pattern that can be used to identify entity attributes of named entity objects in product document information is pre-written. A regular expression pattern can be composed of simple characters, or can be composed of multiple characters and different methods. made.

For example, insurance product information usually has the following expressions: XX insurance, insurance with XX people as the insured, or one of XX insurance types, divided into XX health insurance, education insurance, etc., you can set and The regular expression pattern that matches the above expressions, and then identifies the expression method of the specific attribute content in the insurance product information.

Step S240, using the named entity object, entity attribute and entity relationship corresponding to each product knowledge point as source data to construct a product knowledge graph based on a graph database;

In this embodiment, the named entity object, entity attribute and entity relationship corresponding to each product knowledge point are used as source data to construct a product knowledge structure graph based on a graph database, for example, an insurance product knowledge graph is constructed. At the same time, based on the different types of insurance product knowledge points extracted, different insurance product knowledge structure diagrams are correspondingly constructed.

Step S250, based on the product knowledge graph, create the question set and the product adaptation table.

In this embodiment, based on the difference of main contents in the constructed product knowledge graph, query pages for different product knowledge can be created for the user to retrieve different product knowledge contents. For example, it can provide not only the retrieval of knowledge points of children's health insurance, but also the retrieval of similar insurance types such as children's education insurance, so as to provide users with more comprehensive and efficient information retrieval services.

In this embodiment, the product document information of the designated website is collected through the web crawler technology, and the product document information is processed and arranged to form a product knowledge structure diagram; For multiple rounds of conversations to collect user information. The product document information is collected by crawling, so the collected product document information can meet the query needs of general users for product knowledge. In addition, this embodiment uses the product knowledge structure diagram as the retrieval database of the product query service, which can not only provide massive product document knowledge, but also provide users with a more efficient and fast product knowledge query service, thereby optimizing product configuration.

In this embodiment, the product adaptation table is a collection of insurance products that are more suitable for users to purchase based on multi-dimensional information such as the user's age, work nature, income status (current and expected), family status, risk preference, and financial ability. Adaptation rules, configure suitable products according to users' subjective needs and recommend them to users.

Referring to FIG. 5 , FIG. 5 is a schematic diagram of a refinement flow of an embodiment of step S120 in FIG. 2 . Based on the foregoing embodiment, in this embodiment, the foregoing step S120 further includes:

Step S1201, dividing the user reply information into multiple data sections;

In this embodiment, the user reply information is subjected to analysis processing, the data information generated after the analysis processing is divided into a plurality of data sections, and emotion judgment is performed on each data section respectively. For example, the user's reply information is "two life insurance and children's insurance, which one is more suitable for children to buy, can you recommend it?" It is more suitable to buy for children", "recommended".

Step S1202, performing emotion judgment on the data sections respectively, and obtaining the degree attribute values of various emotions;

In this embodiment, the emotion categories may include, but are not limited to, happy, surprised, normal, angry, bored, sad, and the like. The emotion judgment is performed on each data segment of the user's reply information respectively, and the degree attribute value corresponding to each emotion is obtained. For example, the user's reply information is "two life insurance and children's insurance, which one is more suitable for children to buy, can you recommend it?" Which one is more suitable for children?" and "recommended", and make emotional judgments for each data segment.

Step S1203, based on the degree attribute value of the emotion of the data segment, respectively determine the emotion label of the data segment;

In this embodiment, it is obtained that each data segment may include several emotions, and at the same time, the degree attribute values of various emotions are different, and the corresponding emotion label with the largest degree attribute value is the data segment. emotional labels. For example, it can be obtained that the degree attribute value of the “happy” emotional label is 0.522, the degree attribute value of the “surprise” emotional label is 0.1, the degree attribute value of the “normal” emotional label is 0.3, and the degree attribute value of the “angry” emotional label is 0.3. 0.01, the degree attribute value of the “annoying” emotional label is 0.01, and the degree attribute value of the “sad” emotional label is 0.01. At this time, the degree attribute value of the “happy” emotional label is the largest, then the emotional category of this data segment is happy.

Step S1204, based on the emotional labels of the data sections, output the probability distribution map of the emotional labels corresponding to each data section in the user reply information;

Step S1205, obtaining the user's emotion based on the probability distribution map of the emotion label.

In this embodiment, according to the emotion tags of each data segment, a probability distribution map of emotion tags corresponding to each data segment in the user reply information is output. For example, the emotional labels of each data segment in the user's reply information can be obtained, respectively, the probability of the "happy" emotional label is 0.522, the probability of the "surprise" emotional label is 0.12, the probability of the "normal" emotional label is 0.32, " The probability of the emotional label of "anger" is 0.018, the probability of the emotional label of "annoying" is 0.01, and the probability of the emotional label of "sad" is 0.01. At this time, the probability of each emotional label can be thresholded, and according to the threshold judgment result from The emotion categories of the user are determined from these emotion labels. If the probabilities of all emotion tags are less than the preset threshold, the emotion tag with the highest probability is used as the user emotion output.

In addition, this method can not only identify the user's emotions, but also facilitate the subsequent dialogue interaction process, so that the system can give different emotional feedback to the user. Euphemistic secondary confirmation, so as to obtain the real needs of users and avoid being misled by untrue information.

Referring to FIG. 6 , FIG. 6 is a schematic diagram of a refinement flow of step S130 in FIG. 2 . Based on the foregoing embodiment, in this embodiment, the foregoing step S130 further includes:

Step S1301, analyzing the user reply information in word units to obtain a named entity as a user intent parameter candidate;

In this embodiment, according to the specific data content of the user's reply information, intention analysis is performed in units of words, and a corresponding named entity is obtained, which is used as a parameter candidate of the user's intention. The specific implementation steps are as follows, for example, "buy a critical illness insurance for my mother", in which word segmentation, part-of-speech tagging, and named entity recognition: for /pme/rmum/nbuy/vone/m copies/qcritical illness risk/n, the named entity collection is empty. Obtaining intent parameter candidates: From the results of word segmentation and part-of-speech tagging, it can be seen that a critical illness insurance is a combination of numeral m + quantifier q + noun n, and the noun "critical illness insurance" is attributed to the insurance product intent parameter, then insurance = a critical illness insurance risk.

Step S1302, parse the user reply information, and fuzzy match word by word according to a preset user intent key candidate set to obtain intent keywords;

In this embodiment, the user reply information is parsed, and the intent keywords are obtained by word-by-word fuzzy matching according to a preset user intent key candidate set. For example, "buy a critical illness insurance for my mother", dependency parsing: 0: give/p 3: ADV 1: me/r 2: ATT 2: mother/n 0: POB 3: buy/v-1: HED 4: one/m 5: ATT 5: copies/q 6: ATT 6: critical illness insurance/n 3: VOB.

Obtaining intent keywords: Compare the similarity between the sorted out intent keywords for buying critical illness insurance (for example, "buy a critical illness insurance") and the word segmentation results of the user's reply information, and calculate to get cosine (purchase, a critical illness insurance) insurance) = 1.89, meeting similar requirements, and using "purchase" as the intent keyword.

Step S1303, output the user intent identification result based on the intent keyword;

In this embodiment, the dependency relationship between the user intent parameter candidates is determined according to the obtained intent keywords. From the results of the dependency syntax analysis, it can be seen that there is a dependency between the intent parameter a critical illness insurance and the intent keyword "purchase" (purchase, a critical illness insurance, VOB), and it can be concluded that the user's intention is "purchase critical illness insurance". risk"

Step S1304: Determine the purchase intention of the user based on the user intention identification result and the user emotion.

In this embodiment, based on the user's basic intention information and the user's emotion, the user's deep-level needs, that is, the user's purchase intention, are determined. The basic intent information corresponds to the intent intuitively reflected by the user's reply data, but it cannot reflect the user's real needs in the current state. Therefore, it is necessary to combine the user's emotions to comprehensively determine the real purchase intent that the user's reply information actually wants to express. and emotional needs. For example, "I want to buy insurance, is critical illness insurance more suitable for me, or is life insurance more suitable for me?" and "If I want to buy insurance, is critical illness insurance more suitable for me, and life insurance is also suitable for me?" With different emotions, the real purchasing needs that users want to express are completely different.

Referring to FIG. 7 , FIG. 7 is a schematic diagram of a refinement flow of an embodiment of step S140 in FIG. 2 . Based on the foregoing embodiment, in this embodiment, the foregoing step S140 further includes:

Step S1401, translating the user information into a label format, and generating multiple labels corresponding to the user information;

In this embodiment, the user information is processed according to the preset label format to generate field information in the label format, wherein each generated field information in the label format is a label. Among them, the generated set of multiple tags is called tag set. For example, the user's reply message is "My name is Li Si, gender is female, I am a teacher, I am 43 years old this year, my annual income is 150,000 yuan, and my family assets are about 2 million yuan. I want to buy a critical illness insurance." Assignment, the "gender" label is 2, the "age" label is 6, the "occupation" label is 5, the "annual income" label is 9, and the "family assets" label is 13, and they are translated into label forms, then Li Si = {2,6,5,9,13,}.

Step S1402, if the purchase intention of the user is a positive intention, then based on the preset weight value of the tag, the tags are respectively matched with the products in the preset product adaptation table, and each tag is calculated and output. The matching score of the corresponding product;

In this embodiment, if the user's purchase intention is a positive intention, each tag is matched with multiple insurance products in the product adaptation table according to the weight value of each tag, and each tag and product adaptation table is calculated. Match score for each product in . For example, the user's reply message "My name is Li Si, gender is female, I am a teacher, I am 43 years old this year, my annual income is 150,000 yuan, and my family assets are about 2 million yuan. I want to buy a critical illness insurance" into a label format, Then it is Li Si = {2, 6, 5, 9, 13}, respectively, label 2, label 6, label 5, label 9, label 13 and product A, product B, product C in the preset product adaptation table , product D, and product E are matched, and according to the weight value of each label, the matching score between each label and the corresponding product is calculated and output.

Step S1403, based on the matching score of each label and the corresponding product, summing the matching score of each label and the corresponding product to obtain the matching degree score of each product;

In this embodiment, according to the matching scores of each label and each product, the matching scores of each label are summed to obtain the score of the matching degree of each product in the product adaptation table. For example, the user's reply message "My name is Li Si, gender is female, I am a teacher, I am 43 years old this year, my annual income is 150,000 yuan, and my family assets are about 2 million yuan. I want to buy a critical illness insurance" and translate it into a label. Then Li Si = {2, 6, 5, 9, 13}, respectively label 2, label 6, label 5, label 9, label 13, and match each label with product A in the preset product adaptation table , the matching scores are 5, 7, 2, 11, 7, respectively, and the matching scores of these tags are summed, so the matching degree of product A is divided into 5+7+2+11+7=32, and the presets are obtained in turn Score for the degree of fit of all products in the product fit table.

Step S1404, if the matching degree score of the product is higher than a preset threshold, mark the product as recommended, if not, mark it as non-recommended;

In this embodiment, according to the matching degree score of each product, if the product matching degree score is higher than the preset value, the product is marked as recommended, and if the product matching degree score is lower than the preset value, it is marked as non-recommended. For example, the matching scores of all products in the preset product adaptation table are 35, 47, 55, 55, and 85, respectively. If the matching degree score of the product is higher than the preset threshold, the product is marked as recommended. If not, mark as non-recommended.

In step S1405, the preset number of products marked as recommended are taken as the recommended results, and output.

In this embodiment, a certain number of products marked as recommended are sent to the client as the recommendation results and presented to the user. For example, the matching scores of all products Product A, Product B, Product C, Product D, and Product E in the preset product adaptation table are 35, 47, 55, 55, and 85, respectively. If the threshold is set to 50, product C, product D, and product E are marked as recommended, and product C, product D, and product E are recommended to users.

Referring to FIG. 8 , FIG. 8 is a schematic diagram of functional modules of an embodiment of an intelligent product matching device of the present invention. In this embodiment, the product intelligent matching device includes:

The acquiring module 10 is used for conducting multiple rounds of dialogues with the user based on the preset question set, and acquiring the user's reply information for each round of dialogues, so as to collect the user information;

The emotion recognition module 20 is used for performing emotion recognition on the user reply information to obtain the current user emotion;

an intention recognition module 30, which performs intention recognition on the user's reply information, and determines the user's purchase intention based on the user's emotion;

The product recommendation module 40 is configured to, when the user's intention is a positive intention, query a preset product adaptation table based on the user information, and output a product matching the user information.

Optionally, in a specific embodiment, the product intelligent matching device further includes:

The collection module is used to collect all the product document information in the preset format displayed on the designated website through the web crawler technology;

The building module is used to perform entity relationship extraction processing on the product document information, and obtain the named entity objects of each product knowledge point in the product document information and the entity relationship between the named entity objects; according to the named entity objects of each product knowledge point object, using regular expressions to identify the entity attributes corresponding to each named entity object in the product document information; using the named entity objects, entity attributes and entity relationships corresponding to each product knowledge point as source data to build a graph database-based product knowledge map ;

A creation module is configured to create the question set and the product adaptation table based on the product knowledge graph.

Optionally, in a specific embodiment, the obtaining module includes:

an acquisition unit, used to acquire the user's reply information for each round of dialogue;

a slot matching unit, configured to match the user reply information with a preset slot in a preset question set;

an extraction unit, configured to extract the preset slot if the user reply information is successfully matched with the preset slot;

an association unit, configured to associate the answer information with the preset slot if there is answer information matching the preset slot in the user reply information;

A collection unit, configured to collect user information based on the association result.

Optionally, in a specific embodiment, the emotion recognition module includes:

a dividing unit, configured to divide the user reply information into a plurality of data sections;

a judging unit, which is used to judge the emotions of the data sections respectively, and obtain the degree attribute values of various emotions;

An emotion recognition unit, configured to respectively determine the emotion labels of the data segments based on the degree attribute value of the emotion of each data segment; output the emotion label probability distribution corresponding to each emotion based on the emotion labels of the various data segments Figure; based on the emotion label probability distribution map, the user emotion is obtained.

Optionally, in a specific embodiment, the intent recognition module includes:

an analysis unit, configured to analyze the user reply information in word units to obtain a named entity as a user intent parameter candidate;

an intent matching unit, configured to parse the user reply information, and obtain intent keywords by word-by-word fuzzy matching according to a preset user intent key candidate set;

An intent identification unit, configured to output a user intent identification result based on the intent keyword; and determine the user's purchase intent based on the user intent identification result and the user emotion.

Optionally, in a specific embodiment, the product recommendation module includes:

a format translation unit, configured to translate the user information into a label format, and generate a plurality of labels corresponding to the user information;

The calculation unit is configured to, if the purchase intention of the user is a positive intention, respectively match the tags with the products in the preset product adaptation table based on the preset weight values of the tags, calculate and output each tag. The matching score of the label and the corresponding product; based on the matching score of each label and the corresponding product, the matching score of each label and the corresponding product is summed to obtain the matching degree score of each product;

A product recommendation unit, configured to mark the product as recommended if the matching degree score of the product is higher than a preset threshold, and if not, mark the product as non-recommended; take a preset number of products marked as recommended as recommendations result and output.

The product intelligent matching method proposed in this embodiment improves the current market deficiencies in the so-called intelligent insurance assistant business model and technology, and provides a product intelligent matching method based on user information and user needs. Collect the product document information of the designated website through the web crawler technology, process the product document information, and build a problem set for product knowledge points for multiple rounds of dialogue to collect user information, so as to avoid the possibility of simple positive collection of user information. The resulting lack of information or inconsistent understanding of insurance. At the same time, it collects user personalized information based on multiple dimensions, and solves the problem of user needs and product matching through the knowledge graph question and answer of the preset model. After the user is matched with the product adaptation model, accurate product recommendation can be made.

Based on the description content of the above-mentioned embodiment of the intelligent product matching method of the present invention, the content of the embodiment of the product intelligent matching device will not be described in detail in this embodiment.

The present invention also provides a computer-readable storage medium.

In this embodiment, a product intelligent matching program is stored on the computer-readable storage medium, and when the product intelligent matching program is executed by the processor, the steps of the product intelligent matching method described in any of the above embodiments are implemented. The method implemented when the product intelligent matching program is executed by the processor may refer to the various embodiments of the product intelligent matching method of the present invention, and thus will not be described repeatedly.

From the description of the above embodiments, those skilled in the art can clearly understand that the method of the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is better implementation. Based on this understanding, the technical solutions of the present invention essentially or the parts that contribute to the prior art can be embodied in the form of software products, and the computer software products are stored in a storage medium (such as ROM/RAM), including Several instructions are used to cause a terminal (which may be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention.

The embodiments of the present invention have been described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific embodiments, which are merely illustrative rather than restrictive. Under the inspiration of the present invention, without departing from the scope of protection of the purpose of the present invention and the claims, many forms can be made. Directly or indirectly applied in other related technical fields, these all belong to the protection of the present invention.

Claims (10)

1. a product intelligent matching method, is characterized in that, described product intelligent matching method comprises the following steps: outputting a preset question set to the user, and obtaining the user's response information for each question in the preset question set; Perform emotion recognition on the user's reply information to obtain the user's emotion information; Perform intention identification on the user's reply information to obtain the user's intention information, and determine the user's purchase intention based on the user's emotion information and the user's intention information; If the user's purchase intention is a positive intention, based on the user information, a preset product adaptation table is queried, and a product matching the user information is output. 2 . The intelligent product matching method according to claim 1 , wherein before the outputting the preset question set to the user, and acquiring the user response information of the user to each question in the preset question set, the method further comprises: 2 . : Collect all product document information in the preset format displayed on the specified website through web crawler technology; Performing entity relationship extraction processing on the product document information to obtain named entity objects of each product knowledge point in the product document information and entity relationships between the named entity objects; According to the named entity object of each product knowledge point, using a regular expression to identify the entity attribute corresponding to each named entity object in the product document information; Using the named entity objects, entity attributes and entity relationships corresponding to each product knowledge point as source data, build a product knowledge graph based on graph database; Based on the product knowledge graph, the question set and the product adaptation table are created. 3. The method for intelligent product matching according to claim 1, wherein the outputting a preset question set to a user, and acquiring the user response information of the user to each question in the preset question set comprises: Obtain the user's reply information for each round of dialogue; matching the user reply information with the preset slots in the preset question set; If the user reply information is successfully matched with the preset slot, extract the preset slot; If there is answer information matching the preset slot in the user answer information, associating the answer information with the preset slot; Based on the association result, user information is collected. 4. The method for intelligent product matching according to claim 1, wherein the performing emotion recognition on the user reply information, and obtaining the user emotion information comprises: dividing the user reply information into a plurality of data segments; Perform emotional judgments on the data sections respectively to obtain the degree attribute values of various emotions; Based on the emotional degree attribute value of each data section, determine the emotional label of the data section respectively; Based on the emotion labels of the data sections, outputting the emotion label probability distribution map corresponding to each emotion; The user emotion information is obtained based on the emotion tag probability distribution map. 5 . The intelligent product matching method according to claim 1 , wherein the user's reply information is subjected to intention identification to obtain the user's intention information, and the user's intention information is obtained based on the user's emotion information and the user's intention information. 6 . , and determining the user's purchase intention includes: Analyzing the user reply information in word units to obtain a named entity as a user intent parameter candidate; Analyzing the user reply information, and fuzzy matching word by word according to a preset user intent key candidate set to obtain intent keywords; outputting a user intent identification result based on the intent keyword; The purchase intention of the user is determined based on the user intention recognition result and the user emotion information. 6. The intelligent product matching method according to any one of claims 1-5, wherein, if the user's purchase intention is a positive intention, query preset product adaptation based on the user information Table, the output products that match the user information include: Translate the user information into a label format, and generate a plurality of labels corresponding to the user information; If the purchase intention of the user is a positive intention, then based on the preset weight value of the tag, each tag is matched with the product in the preset product adaptation table, and the relationship between each tag and the corresponding product is calculated and output. match score; Based on the matching score of each label and the corresponding product, summing the matching score of each label and the corresponding product to obtain the matching degree score of each product; If the matching degree score of the product is higher than the preset threshold, mark the product as recommended, if not, mark it as non-recommended; The preset number of products marked as recommended are output as recommended results. 7. A product intelligent matching device, wherein the product intelligent matching device comprises: The acquisition module is used to conduct multiple rounds of dialogue with the user based on the preset question set, and acquire the user's reply information for each round of dialogue to collect user information; an emotion recognition module, configured to perform emotion recognition on the user response information to obtain current user emotion information; an intention recognition module, which performs intention recognition on the user's reply information, and determines the user's purchase intention based on the user's emotion; The product matching module is used for querying a preset product adaptation table based on the user information when the user's intention is a positive intention, and outputting a product matching the user information. 8. The product intelligent matching device according to claim 7, wherein the acquisition module comprises: an acquisition unit, used to acquire the user's reply information for each round of dialogue; a matching unit, configured to match the user reply information with a preset slot in a preset question set; an extraction unit, configured to extract the preset slot if the user reply information is successfully matched with the preset slot; an association unit, configured to associate the answer information with the preset slot if there is answer information matching the preset slot in the user reply information; A collection unit, configured to collect user information based on the association result. 9. A product intelligent matching device, characterized in that, the product intelligent matching device comprises a memory, a processor, and a product intelligent matching program stored on the memory and running on the processor, and the product intelligent matching program When the matching program is executed by the processor, the steps of the intelligent product matching method according to any one of claims 1-6 are realized. 10. A computer-readable storage medium, characterized in that a product intelligent matching program is stored on the computer-readable storage medium, and when the product intelligent matching program is executed by a processor, any one of claims 1-6 is realized. The steps of the product intelligent matching method described in the item.

Images