KR102594734B1 - Text analysis method using lda topic modeling technique and text analysis apparatus performing the same - Google Patents

Abstract

The present invention is a text analysis method for solving the first above-described problem through an LDA model provided in the device, and provides a method for analyzing text that corresponds to at least one topic included in source data through an LDA model provided in the device. Obtaining a language output value including unique identification information and a probability of occurrence of the topic;
counting the number of occurrences of the topic corresponding to each of the language output values; extracting the probability of occurrence from each of the language output values; calculating a probability of occurrence of the language output value relative to the number of occurrences of the topic; and
It may include determining the semantic recognition rate and the structure recognition rate as at least one of richness, clarity, and noise of the source data determined based on the language output value.

Description

Text analysis method using LDA (Latent Dirichlet Allocation) topic modeling technique and text analysis device for performing the same {TEXT ANALYSIS METHOD USING LDA TOPIC MODELING TECHNIQUE AND TEXT ANALYSIS APPARATUS PERFORMING THE SAME}

The present invention relates to a text analysis method, and specifically to a text analysis method and text analysis device using the LDA (Latent Dirichlet Allocation) topic modeling technique.

Natural Language Processing is an element technology that includes natural language analysis, understanding, and generation, and is applied to various fields such as information retrieval, machine translation, and question and answering.

Natural language processing uses technologies such as natural language analysis, natural language understanding, and natural language generation. Natural language analysis can be divided into four types according to its degree: morphological analysis, syntactic analysis, semantic analysis, and pragmatic analysis. Natural language understanding is a technology that allows a computer to operate according to input given in natural language, and natural language generation is a technology that converts the contents of videos or tables into natural language that humans can understand.

Recently, a neural network model has been used in natural language processing.

Although these neural network models provide improved performance in semantic analysis in natural language processing, they have the problem of not providing high accuracy and operating with inconsistent behavior when the source data is small.

Public Patent Publication No. 10-2019-0046631

The problem to be solved by the present invention is to provide a text analysis method and text analysis device that perform natural language processing with high accuracy and consistency even when using a small amount of source data by using a neural network model and a non-neural network model together.

The problems to be solved by the present invention are not limited to the problems mentioned above, and other problems not mentioned can be clearly understood by those skilled in the art from the description below.

The present invention to solve the above-described problem is a text analysis method, a language including unique identification information corresponding to at least one topic included in source data and a probability of occurrence of the topic through an LDA model provided in the device. Obtaining an output value;

counting the number of occurrences of the topic corresponding to each of the language output values; extracting the probability of occurrence from each of the language output values; calculating a probability of occurrence of the language output value relative to the number of occurrences of the topic; and determining a semantic recognition rate corresponding to the at least one topic and a structure recognition rate of the source data based on the language output value for the number of occurrences of the topic.

At this time, the step of determining the semantic recognition rate and the structural recognition rate is,

It may include determining the semantic recognition rate and the structure recognition rate as at least one of richness, clarity, and noise of the source data determined based on the language output value.

In addition, the step of determining the semantic recognition rate corresponding to at least one topic and the structural recognition rate of the source data includes:

The probability of occurrence of the language output value with respect to the number of occurrences of the topic is integrated with the number of occurrences of the topic to calculate the richness of the source data, and the meaning corresponding to the topic based on the richness of the source data. It may include determining a recognition rate and a structure recognition rate of the source data.

Additionally, in the step of determining the recognition rate corresponding to at least one topic and the structure recognition rate of the source data, the semantic recognition rate and the structure recognition rate may be determined to decrease as the amount of meaning of the source data increases.

In addition, the step of determining the recognition rate corresponding to at least one topic and the structure recognition rate of the source data determines the semantic clarity of the source data based on the amount of change in the probability of occurrence of the language output value with respect to the number of occurrences of the topic. It may include calculating and determining the semantic recognition rate and the structure recognition rate of the source data based on the semantic clarity of the source data.

In addition, the step of determining the semantic recognition rate and the structural recognition rate is,

As the semantic clarity increases, the semantic recognition rate and the structural recognition rate may be determined to increase.

Additionally, the step of determining a recognition rate corresponding to at least one subject and a structure recognition rate of the source data includes:

Calculating an average value of the number of topic occurrences, calculating noise of the source data based on a change in the probability of occurrence of the language output value for the number of topic occurrences corresponding to less than or equal to the average number of topic occurrences,

It may include determining the semantic recognition rate and the structure recognition rate of the source data based on noise of the source data.

Additionally, in the step of determining the semantic recognition rate and the structural recognition rate, the semantic recognition rate and the structural recognition rate may be determined to decrease as noise of the source data increases.

Additionally, in the step of acquiring the language output value, the language output value may be obtained by extracting the at least one topic independently of the order in the source data.

The text analysis method according to an embodiment of the present invention performs natural language recognition of the source data based on the semantic recognition rate and the structure recognition rate, extracts advanced semantic features, and uses the extracted advanced semantic features. It may further include performing learning of a non-neural network model.

A text analysis device according to an embodiment of the present invention includes a device for performing natural language processing, a memory configured to store source data and an LDA model; and

At least one processor configured to communicate with the memory,

The at least one processor obtains a language output value including unique identification information and an occurrence probability corresponding to at least one subject included in the source data from an LDA model previously formed using source data, and outputs the language. Count the number of occurrences of the topic corresponding to each value, extract the probability of occurrence from each of the language output values, calculate the probability of occurrence of the language output value for the number of occurrences of the topic, and calculate the probability of occurrence of the language output value for the number of occurrences of the topic. The semantic recognition rate and the structural recognition rate may be determined by at least one of the richness, clarity, and noise of the source data calculated based on the language output value.

Other specific details of the invention are included in the detailed description and drawings.

The text analysis method and text analysis device according to an embodiment of the present invention can perform natural language processing with high accuracy and consistency even when using a small amount of source data by using a neural network model and a non-neural network model together.

The text analysis method and text analysis device according to an embodiment of the present invention utilize a statistical method and an LDA (Latent Dirichlet allocation) topic modeling technique in semantic analysis to derive the recognition difficulty of text, thereby efficiently Natural language processing can be performed.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

1 is a block diagram of a text analysis device according to an embodiment of the present invention.
Figures 2 and 3 are diagrams for explaining the operation of a neural network model and a non-neural network model according to an embodiment of the present invention.
Figure 4 is a diagram for explaining a neural network model according to an embodiment of the present invention.
5A and 5B are diagrams for explaining advanced semantic analysis according to an embodiment of the present invention.
6 to 10 are flowcharts according to an embodiment of the present invention.

The advantages and features of the present invention and methods for achieving them will become clear by referring to the embodiments described in detail below along with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various different forms. The present embodiments are merely provided to ensure that the disclosure of the present invention is complete and to provide a general understanding of the technical field to which the present invention pertains. It is provided to fully inform the skilled person of the scope of the present invention, and the present invention is only defined by the scope of the claims.

The terminology used herein is for describing embodiments and is not intended to limit the invention. As used herein, singular forms also include plural forms, unless specifically stated otherwise in the context. As used in the specification, “comprises” and/or “comprising” does not exclude the presence or addition of one or more other elements in addition to the mentioned elements. Like reference numerals refer to like elements throughout the specification, and “and/or” includes each and every combination of one or more of the referenced elements. Although “first”, “second”, etc. are used to describe various components, these components are of course not limited by these terms. These terms are merely used to distinguish one component from another. Therefore, it goes without saying that the first component mentioned below may also be a second component within the technical spirit of the present invention.

Unless otherwise defined, all terms (including technical and scientific terms) used in this specification may be used with meanings commonly understood by those skilled in the art to which the present invention pertains. Additionally, terms defined in commonly used dictionaries are not interpreted ideally or excessively unless clearly specifically defined.

In this specification, source data may refer to data including text data used in natural language processing.

In this specification, advanced semantic features may refer to features extracted through performing semantic analysis of source data.

In this specification, a hybrid model may refer to a natural language processing model that uses both a neural network model and a non-neural network model.

In this specification, language output value may refer to output data whose occurrence probability matches identification information corresponding to a topic extracted from source data.

In this specification, the meaning recognition rate may refer to the rate at which a text analysis device recognizes the meaning of text included in source data.

In this specification, the structure recognition rate may refer to the structure of paragraphs formed by text in source data.

In this specification, the transformation feature may refer to the characteristics of the form of the language in which the text used in the source data is transformed from the language in which it is commonly used.

In this specification, semantic richness may refer to an indicator corresponding to the amount of meaningful text in the text included in the source data.

In this specification, semantic clarity may refer to the degree to which each subject of the text included in the source data is distinct.

In this specification, noise of source data may refer to the ratio of text corresponding to unnecessary meaning in text included in the source data.

Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

Figure 1 is a control block diagram of a text analysis device 10 according to an embodiment of the present invention.

Referring to FIG. 1, a text analysis device according to an embodiment of the present invention may include a memory, at least one processor, and a communication unit.

Meanwhile, the text analysis device can receive multiple source data (D11, D12, D13).

The first source data D11 may mean a database containing advanced meaning.

Additionally, the second source data D12 may mean a database containing general text.

Additionally, the third source data D13 may mean a database modified from a typical language.

The text analysis device can perform both learning of a non-neural network model and learning of a neural network model using data included in the third source data D13.

The communication unit 130 may receive source data from a database.

The communication unit 130 may include one or more components that enable communication with an external device, and may include, for example, at least one of a short-range communication module, a wired communication module, and a wireless communication module.

The memory 110 can store various data for natural language processing, including source data D11, D12, and D13.

The memory 110 includes non-volatile memory elements such as cache, read only memory (ROM), programmable ROM (PROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), and flash memory, or RAM ( It may be implemented as at least one of a volatile memory device such as Random Access Memory (Random Access Memory) or a storage medium such as a hard disk drive (HDD) or CD-ROM, but is not limited thereto.

At least one processor 120 may include a natural language recognition unit 121, a neural network model learning unit 122, and a non-neural network model learning unit 123.

Additionally, at least one processor 120 may include a central processor 124 to perform the operations of the modules described above.

In addition, at least one processor 120 may include a central processor 124 capable of performing the operations of the natural language recognition unit 121, the neural network model learning unit 122, and the non-neural network model learning unit 123 described above. there is.

At least one processor 120 may perform natural language recognition of the first source data D11 through a pre-formed LDA model to extract advanced semantic features.

The LDA model may refer to latent Dirichlet allocation (LDA).

The advanced semantic feature may refer to the semantic feature of the text corresponding to the first source data D11.

LDA can refer to one of the probabilistic topic model techniques for describing what topics exist in each document for a given document. A detailed explanation related to LDA is provided below.

At least one processor 120 may extract natural language features including advanced semantic features using the second source data D12.

That is, the second source data (D12) is data used to form learning data required for typical natural language processing, and includes not only the advanced semantic features processed with the first source data (D11), but also discourse-based features, syntactic features, transformation features, and Natural language features, including surface features, can be extracted. A detailed explanation related to this will be provided later.

Additionally, at least one processor 120 may perform non-neural network model training for natural language recognition using advanced semantic features and natural language features.

On the other hand, at least one processor 120 may perform neural network model learning for natural language recognition using the second source data D12.

However, at least one processor 120 uses a database of third source data D13 containing a large number of modified texts to additionally extract natural language features including modified features (lxSem), and uses this to create a non-neural network model. Learning can be done.

At least one processor 120 may perform additional learning of the non-neural network model using result data learned by the neural network model.

At least one processor 120 may then form a hybrid model for natural language recognition using both the non-neural network model and the neural network model. A detailed description related to this will be provided in FIG. 2.

Meanwhile, at least one processor 120 may obtain a language output value including unique identification information and an occurrence probability corresponding to at least one subject included in the source data from a pre-formed LDA model using the source data.

The unique identification number may refer to identification information corresponding to a meaningful unit among text included in source data.

Additionally, the probability of occurrence may refer to the probability of occurrence of text in each meaningful unit.

At least one processor 120 may count the number of occurrences of a topic corresponding to each language output value.

Additionally, at least one processor 120 may extract only the probability of occurrence from each language output value and then calculate the probability of occurrence of the language output value based on the number of occurrences of the topic.

At least one processor 120 may determine a semantic recognition rate and a structure recognition rate of source data corresponding to the at least one topic based on the language output value for the number of topic occurrences.

Detailed operations of each calculation by the text analysis device 10 will be described below.

At least one component may be added or deleted in accordance with the performance of the components of the text analysis device 10 shown in FIG. 1. Additionally, it will be easily understood by those skilled in the art that the mutual positions of the components may be changed in response to the performance or structure of the system.

Meanwhile, each component shown in FIG. 1 refers to software and/or hardware components such as Field Programmable Gate Array (FPGA) and Application Specific Integrated Circuit (ASIC).

Figures 2 and 3 are diagrams for explaining the operation of a neural network model and a non-neural network model according to an embodiment of the present invention.

Referring to Figure 2, the text analysis device can perform neural network learning and non-neural network learning using first source data (D21), second source data (D23), and third source data (D22).

Specifically, the text analysis device can extract advanced semantic features (F21) using the first source data (D21) through the LDA technique (M21). A text analysis device can extract the richness, clarity, noise, and total number of discovered topics of each text from a database such as Wikipedia.

Additionally, the text analysis device may extract natural language features including discourse-based features (Disco, F22) that include the high-level structure of natural language sentences included in the second source data (D23).

Discourse-based features (F22) may contain higher-level dependency structures.

Discourse-based features (F22) may include trending features of text structure at micro and macro levels. Discourse-based features (F22) may include entity density function (EnDF) and entity grid (EnGF) functions.

The entity density feature is related to the degree of recognition based on the number of entities, and the entity grid feature is related to the consistency of the corresponding syntax.

At least one processor may extract natural language features including features of phrases and clauses constituting the natural language included in the second source data (D23) and syntax features (Synta, F24) corresponding to the structures of the phrases and clauses.

The syntactic feature (F24) is associated with longer processing times of text.

These syntactic features (F24) can affect the overall complexity of the text, which is an important indicator of readability.

The syntactic feature (F24) can implement several variants, including the number of nouns, verbs, and adverbial phrases.

Additionally, syntactic features (F24) may include the structural form of the parsed tree, working on the average parsing tree height.

Additionally, at least one processor may extract natural language features including transformation features (lxSem, F23) corresponding to transformational use of nouns, verbs, adjectives, and adverbs of natural language included in the second source data (D23).

Modification features (F23) may include properties related to the meaning of the vocabulary included in the source data, the difficulty of the word, and the unfamiliarity of the word.

Specifically, the transformation feature (F23) may include the degree of transformation of nouns, verbs, adjectives, and adverbs that represent the ratio of each word.

Meanwhile, this modified feature (F23) can also be extracted using pre-formed third source data (D22).

Additionally, at least one processor may perform learning of the non-neural network model (M24) using deformed features extracted through the third source data (D22).

In other words, the processor can perform non-neural network learning (M24) using a database other than the database used for general learning and then use it to perform natural language processing.

Additionally, the processor may extract natural language features including surface features (ShaTr, F25) corresponding to syllable features of the natural language included in the second source data (D23).

The surface feature (F25) may refer to a feature corresponding to the level of difficulty.

Surface features (F25) may include the average number of tokens, syllables, and characters of text included in the source data.

Meanwhile, referring to FIGS. 2 and 3 together, at least one processor may perform neural network learning (M22, M23) using the second source data (D23), and the first source data (D21) and the second source Non-neural network learning (M24) may be performed using the data (D23) and the third source data (D32).

Meanwhile, the processor can use both of these models to form a hybrid model.

Referring to FIG. 3, the processor calculates a first weight (W1) and a second weight (W1) corresponding to each of the non-neural network model (M31) and the neural network model (M32) based on the amount of text included in the first source data and the second source data. The weight (W2) can be determined.

In general, neural network models (M31) perform poorly on small data sets, and when source data is small, there may be a lack of consistency depending on each corpus classification method.

Therefore, if the amount of source data is small, learning is performed by increasing the weight (W1) corresponding to the non-neural network model (M31) to enable smooth natural language processing even when the source data is small, and if the amount of source data is large, neural network The performance of natural language processing can be improved by increasing the weight corresponding to the model (M21).

The processor forms a hybrid model (M33) by reflecting the result data output by the non-neural network model (M31) and the result data output by the neural network model (M32) based on the first weight (W1) and the second weight (W2). can do.

The hybrid model (M33) formed in this way can perform effective and consistent natural language processing even with a small amount of source data.

In summary, the processor can form a non-neural network learning model (M31) by performing non-neural network learning using first source data, second source data, and third source data.

Additionally, the processor may form a neural network model (M32) by performing neural network learning using the second source data.

On the other hand, if the amount of source data required for learning is large, the processor can perform learning primarily on the neural network model (M32). If the amount of source data required for learning is small, the processor can perform learning primarily on the non-neural network model (M31). This proportion corresponds to each model. This can be achieved by changing the weights (W1, W2).

Meanwhile, the operation of the present invention described in FIGS. 2 and 3 is only an embodiment of the present invention, and there is no limitation to the embodiment in which the processor forms a hybrid model to improve natural language processing capabilities.

Figure 4 is a diagram for explaining a neural network model using RNN according to an embodiment of the present invention.

In general, a recurrent neural network (RNN) can be used in natural language processing.

Recurrent neural network (RNN) is a type of artificial neural network, and is characterized by a circular structure in the connections between units.

This structure allows the state to be stored inside the neural network so that time-varying dynamic features can be modeled, so unlike a forward neural network, input in the form of a sequence can be processed using internal memory.

Therefore, recurrent artificial neural networks can be applied to process data with time-varying characteristics, such as natural language recognition. Because of these characteristics, RNNs can be used by processors to perform natural language processing.

In other words, the processor can use a neural network model using RNN to perform learning through a neural network model.

Specifically, the RNN used by the processor can learn the sequence information of texts included in the source data by obtaining continuous sequence information rather than simply connecting vectors of tokens (T41, T42, T43, T4n).

However, because the RNN concept uses only the value of the previous cell as input, information loss may gradually occur even if the value of the previous cell includes the calculation result of the previous cell.

Therefore, when the processor performs learning using RNN, it has high accuracy for short sentence structures, but does not have accuracy for long sentence structures.

To overcome this, a large amount of source data learning is required, and smooth natural language processing is difficult when a small amount of source data is learned.

Therefore, to overcome these shortcomings, a hybrid model can be formed using a neural network model using RNN and a non-neural network model that performs learning in parallel.

As described above, the processor may perform non-neural network learning in parallel with a model performing RNN neural network learning and fuse them to form a final hybrid model.

Meanwhile, the operation of the present invention described in FIG. 4 is only an example of an RNN used in natural language processing, and in addition to RNN, Long Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) can be used in natural language processing.

5A and 5B are diagrams for explaining advanced semantic analysis according to an embodiment of the present invention.

Referring to FIG. 5A, the processor may extract at least one topic included in the source data from a pre-formed LDA model using the source data.

LDA (Latent Dirichlet allocation) can refer to a probabilistic generation model for discrete data.

LDA can be used on text-based data.

LDA can refer to topic modeling performed through latent semantic indexing (LSI) and probabilistic latent semantic analysis (pLSA).

LDA makes several assumptions, an important one of which is the exchangeability of words.

Commutativity can mean the assumption that the order of words does not matter, only the presence or absence of words matters.

Therefore, in LDA, if the order of words is ignored, the document can be expressed simply with the frequency of the words included in it.

Based on this assumption, a mixed model that includes the interchangeability of words and documents can be presented through LDA. However, the exchangeability assumption of LDA can be expanded by thinking of a bundle of specific words as a unit (n-gram) rather than simply thinking of a single word as a unit.

That is, the processor of the text analysis device can obtain a language output value by extracting at least one topic independently of the order in the source data.

The text analysis device can match a unique identification number and probability of occurrence to each topic and form a language output value based on this.

In Figure 5a, the first topic was matched with the identification number 3 and the occurrence probability of 0.45, and the second topic was matched with the identification number 7 and the occurrence probability of 0.25.

In addition, the third topic was matched with identification number 42 and an occurrence probability of 0.2, and the last topic was matched with identification number 45 and an occurrence probability of 0.1 (O51).

The processor also derives the probability of occurrence of these topics, but can also count the number of topic occurrences corresponding to how many topics appear in the actual source data.

The processor can output a language output value whose identification number and probability of occurrence are matched to each topic.

Subsequently, the processor can extract only the probability of occurrence of each topic from the language output value (S51).

The processor may calculate the probability of occurrence of the language output value relative to the number of occurrences of the topic and derive the relationship between the number of occurrences of the topic and the probability of occurrence of the topic (R51).

Meanwhile, Figure 5b graphs the relationship between the number of topic occurrences and the probability of topic occurrence.

5B, the processor may determine a semantic recognition rate and a structure recognition rate of source data corresponding to at least one topic based on language output values for the number of topic occurrences, as will be described later.

As described above, the semantic recognition rate may refer to the degree to which a text analysis device recognizes the text itself, and the structural recognition rate may refer to the degree to which the text analysis device recognizes the syntax of the text itself.

First, the processor may calculate the semantic quantity (R62) of the source data by integrating the probability of occurrence of the language output value with respect to the number of occurrences of the topic.

The meaningful quantity (R62) of the source data can be calculated based on Equation 1 below.

[Equation 1]

Referring to Equation 1, S _R may mean the meaningful amount (R52) of the source data.

The amount of meaning may refer to the amount of text that has meaning in the text included in the source data.

p _i may mean the probability of topic occurrence corresponding to each topic, and i may mean the number of topic occurrences, which may correspond to the x-coordinate of the graph in FIG. 5B.

Specifically, in the graph presented in FIG. 5B, the base area may represent the meaningful amount (R62) of the source data.

This large amount of meaning (R52) means that there is a large number of texts to be recognized.

As the semantic amount of source data increases, the semantic recognition rate and the structure recognition rate may be determined to decrease.

In other words, if the amount of meaning is large, the meaning recognition rate and structure recognition rate of the corresponding source data may be determined to be low, and the difficulty of the corresponding source data may be determined to be high and readability to be low.

Additionally, the processor can calculate the semantic clarity (C52) of the source data based on the change in the probability of occurrence of the language output value relative to the number of occurrences of the topic.

The semantic clarity (C52) of the source data can be determined based on the equation below.

[Equation 2]

Referring to Equation 2, S _c means clarity of meaning, max(p) means the maximum value of the probability of occurrence of each topic, and pi may mean the probability of topic occurrence corresponding to the number of occurrences of each topic.

That is, the processor can determine semantic clarity (C52) based on the change in probability of occurrence for the number of occurrences of each topic relative to the maximum value of probability of occurrence.

A high semantic clarity (C52) means that the meaning of each text is clearly distinguished, so it can mean that the meaning recognition rate or structure recognition rate is high.

That is, the processor may determine that there is a positive correlation in which the semantic recognition rate and the structural recognition rate increase as semantic clarity increases.

Meanwhile, the processor may further consider noise (N52) in determining the subject recognition rate and structure recognition rate of the source data.

Specifically, the processor calculates the average value of the number of topic occurrences, and calculates the noise (N52) of the source data based on the change in the probability of occurrence of the language output value for the number of topic occurrences corresponding to the average number of topic occurrences. .

The processor can determine the noise (N52) of the source data based on Equation 3 below.

[Equation 3]

Referring to Equation 3, S _N refers to the noise (N52) of the source data. may mean the average of the probability of occurrence of each topic.

Since the noise may be large, it may mean that it is difficult to recognize meaning from the source data, so the processor may determine that the semantic recognition rate and structure recognition rate decrease as the noise (N52) of the source data increases.

Through the above-described operations, the processor calculates the amount of meaning (C52), clarity (C52), and noise (N52) of the source data, and based on this, calculates the semantic recognition rate of the source data or the structural recognition rate of the phrases that make up the source data. You can.

The processor can calculate the semantic recognition rate using the equation below.

[Equation 4]

Referring to Equation 4, R _m refers to the semantic recognition rate, S _c refers to semantic clarity, S _R refers to the amount of meaning, and S _N may refer to noise. k1 is a constant and can be determined in advance based on the type and amount of source data.

The processor can calculate the structure recognition rate based on the equation below.

[Equation 5]

Referring to Equation 5, R _S refers to the structure recognition rate, S _c refers to semantic clarity, S _R refers to the amount of meaning, and S _N may refer to noise. k2 is a constant and can be determined in advance based on the type and amount of source data.

Meanwhile, calculating the above-mentioned semantic quantity (C52), clarity (C52), and noise (N52) with language output values output using the LDA model is only an embodiment of the present invention, and based on this, the semantic recognition rate and structure are There are no restrictions on the operation that determines the recognition rate.

6 to 10 are flowcharts according to an embodiment of the present invention.

Referring to FIG. 6, the text analysis device can acquire source data (S601).

Afterwards, the text analysis device can extract advanced semantic features through LDA (S602).

Additionally, the text analysis device can extract various natural language features in addition to advanced semantic features (S603).

Meanwhile, based on the data derived in this way, the text analysis device can perform learning of a non-neural network model (S604).

Meanwhile, the text analysis device that has acquired the source data can perform learning of the neural network model (S605).

Afterwards, the text analysis device can form a hybrid model using the untrained non-neural network model and the neural network model (S606).

Next, referring to FIG. 7, the text analysis device can acquire source data (S701).

The text analysis device can extract advanced semantic features from the source data (S702).

Discourse-based features can be extracted (S703). Additionally, the text analysis device can extract syntactic features (S704).

Additionally, deformed features can be extracted from source data and surface features can be extracted (S705, S706). In this way, the natural language processing device can perform learning of a non-neural network model based on the characteristics of the extracted data (S707).

Next, Figure 8 is a flowchart for explaining the features of the neural network model and non-neural network model described above.

Referring to FIG. 8, the text analysis device can acquire source data (S801). Such source data can be used for learning through a neural network model or through a non-neural network model. On the other hand, when learning is performed through a neural network model (S802), a large amount of source data is required, the consistency of recognition for each corpus is low, and explanation of recognition is often difficult.

Therefore, the text analysis device can perform non-neural network model learning and neural network model learning in parallel (S803).

Through this operation, a hybrid model can be formed by harmonizing the neural network model and the non-neural network model (S804).

Figure 9 is a flowchart showing the operation of determining the recognition rate of high-level semantics by determining high-level semantic features in a non-neural network model.

Referring to FIG. 9, the text analysis device can acquire source data (S901).

Afterwards, the natural language processing device can obtain a language output value including the topic, identification information, and probability of occurrence (S902).

Meanwhile, the text analysis device can count the number of occurrences of the topic (S903). \

Additionally, the probability of occurrence corresponding to the number of occurrences of each topic can be extracted (S904). Afterwards, the richness, clarity, and noise corresponding to each topic can be determined (S905).

Finally, the text analysis device can determine the meaning recognition rate and structure recognition rate of the corresponding source data (S906).

Referring to FIG. 10, the text analysis device can calculate semantic amount (S1002), clarity (S1003), and noise (S1004) to determine advanced semantic features using source data (S1001). The text analysis device may determine the subject recognition rate and structure recognition rate of the source data (S1005). The text analysis device may perform non-neural network learning (S1006).

The steps of the method or algorithm described in connection with embodiments of the present invention may be implemented directly in hardware, implemented as a software module executed by hardware, or a combination thereof. The software module may be RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), Flash Memory, hard disk, removable disk, CD-ROM, or It may reside on any type of computer-readable recording medium well known in the art to which the present invention pertains.

Above, embodiments of the present invention have been described with reference to the attached drawings, but those skilled in the art will understand that the present invention can be implemented in other specific forms without changing its technical idea or essential features. You will be able to understand it. Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive.

10: Text analysis device
110: memory
120: processor
130: Department of Communications

Claims (9)

In a text analysis method performed by a device,
Obtaining a language output value including unique identification information corresponding to at least one topic included in source data and a probability of occurrence of the topic through an LDA model provided in the device;
counting the number of occurrences of the topic corresponding to each of the language output values;
extracting the probability of occurrence from each of the language output values;
calculating a probability of occurrence of the language output value relative to the number of occurrences of the topic; and
A step of determining a semantic recognition rate corresponding to the at least one topic and a structure recognition rate of the source data based on the language output value for the number of occurrences of the topic,
The step of determining the semantic recognition rate and the structural recognition rate is,
Obtaining a language output value including unique identification information corresponding to at least one topic included in the source data and a probability of occurrence of the topic;
Determine the amount of meaning, semantic clarity, and noise based on the probability of occurrence of the language output value relative to the number of occurrences of the topic,
A step of determining the semantic recognition rate corresponding to the source data based on the semantic amount and the noise,
A text analysis method comprising: determining the structure recognition rate of the source data based on the semantic clarity and the noise.
According to claim 1,
The step of determining a semantic recognition rate corresponding to the at least one topic and a structural recognition rate of the source data includes:
Calculate the richness of the source data by integrating the probability of occurrence of the language output value with respect to the number of occurrences of the topic with respect to the number of occurrences of the topic,
A text analysis method including; determining a semantic recognition rate corresponding to the topic and a structure recognition rate of the source data based on the amount of meaning of the source data.
According to paragraph 2,
The step of determining a recognition rate corresponding to the at least one topic and a structure recognition rate of the source data includes:
A text analysis method wherein the semantic recognition rate and the structural recognition rate are determined to decrease as the semantic amount of the source data increases.
According to claim 1,
The step of determining a recognition rate corresponding to the at least one topic and a structure recognition rate of the source data includes:
Calculate semantic clarity of the source data based on the change in probability of occurrence of the language output value with respect to the number of occurrences of the topic,
A text analysis method including; determining the semantic recognition rate and the structure recognition rate of the source data based on the semantic clarity of the source data.
According to paragraph 4,
The step of determining the semantic recognition rate and the structural recognition rate is,
A text analysis method for determining that the semantic recognition rate and the structural recognition rate increase as the semantic clarity increases.
According to claim 1,
The step of determining a recognition rate corresponding to the at least one topic and a structure recognition rate of the source data includes:
Calculate the average value of the number of occurrences of the topic,
Calculate noise of the source data based on the amount of change in the probability of occurrence of the language output value for the number of topic occurrences corresponding to less than the average value of the number of topic occurrences,
A text analysis method comprising: determining the semantic recognition rate and the structure recognition rate of the source data based on noise of the source data.
According to clause 6,
The step of determining the semantic recognition rate and the structural recognition rate is,
A text analysis method for determining that the semantic recognition rate and the structural recognition rate decrease as noise of the source data increases.
According to paragraph 1,
The step of obtaining the language output value is,
Extracting the at least one topic independently from the order in the source data to obtain the language output value.
In a device that performs natural language processing,
A memory configured to store source data and LDA model; and
At least one processor configured to communicate with the memory,
The at least one processor,
Obtaining a language output value including unique identification information and an occurrence probability corresponding to at least one topic included in the source data from an LDA model formed in advance using source data,
Count the number of occurrences of the topic corresponding to each of the language output values,
Extracting the probability of occurrence from each of the language output values,
Calculate the probability of occurrence of the language output value for the number of occurrences of the topic,
Obtaining a language output value including unique identification information corresponding to at least one topic included in the source data and a probability of occurrence of the topic,
Determine the amount of meaning, semantic clarity, and noise based on the probability of occurrence of the language output value relative to the number of occurrences of the topic,
Determining a semantic recognition rate corresponding to the source data based on the semantic amount and the noise,
A text analysis device that determines a structure recognition rate of the source data based on the semantic clarity and the noise.