KR20210098135A - Apparatus, method and computer program for analyzing query data - Google Patents

Abstract

Provided is a query analysis device for analyzing query data, which includes: an input unit for receiving query data; a natural language processing unit for deriving a first entity name pattern by performing natural language processing on the received query data; a deep learning model processing unit for deriving a second entity name pattern using a deep learning model with respect to the received query data; a similar meta extraction unit for extracting similar metadata for each entity name included in the second entity name pattern; a query analysis unit for performing query analysis on the received query data based on the first entity name pattern, the second entity name pattern, and the similar metadata; and an output unit for outputting a query analysis result for the received query data.

Description

Query analysis apparatus, method and computer program for analyzing query data {APPARATUS, METHOD AND COMPUTER PROGRAM FOR ANALYZING QUERY DATA}

The present invention relates to a query analysis apparatus, method and computer program for analyzing query data.

A question-and-answer system is a system that obtains a response to a given question, and various techniques such as query analysis, information retrieval, and information extraction are used to obtain an appropriate response to a user's query. A user's query may be input as natural language data in the form of text or voice, and may include various pieces of information at the same time. Therefore, it is necessary to analyze the query before searching and extracting information in the question and answer system.

Conventionally, when a user directly inputs a search term such as a movie title, a program name, a genre, or a cast to search for content, a method in which a keyword search result is provided is generally used. However, as a user interface capable of inputting a query is variously changed from a keyboard and a remote control to a voice recognition device, a technology capable of processing a complex search word is required.

A user's query to search for content includes information on the type of media the user is looking for and information restricting the search results using metadata. In this case, query analysis is usually done in two steps. First, the type of media the user is looking for is recognized, and then the metadata included in the user's question is recognized.

Meta data is recognized using an entity name recognizer. Named-entity recognition (NER) is a method of extracting and classifying words by recognizing a predefined classification such as a person's name, group, place, time, unit, etc. means that

The conventional content search method can process a query including the title of the content and basic metadata (genre, cast, etc.), but has limitations in processing a query including various kinds of metadata. A query that includes multiple metadata means, for example, "American movies set in France" or "Fun Korean movies set in a police academy in 2017".

In addition, the conventional content search method has a problem in that the accuracy of the analysis result is deteriorated even when the query includes metadata not defined in advance. Meta data that is not defined in advance may have a similar meaning to predefined metadata but have a different shape, or a case in which a similar meaning does not exist among predefined metadata. As voice recognition technology is used together for content search, queries including metadata that are not defined in advance are increasing.

Korean Patent Publication No. 10-1873873 (published on June 27, 2018)

An object of the present invention is to provide a query analysis apparatus, method, and computer program for processing a query including various metadata or metadata not defined in advance.

An object of the present invention is to provide a query analysis apparatus, method, and computer program for analyzing query data based on natural language processing and deep learning algorithms.

An object of the present invention is to provide a query analysis apparatus, method, and computer program for analyzing query data including a plurality of keywords.

An object of the present invention is to provide a query analysis apparatus, method, and computer program for analyzing query data including keywords that are not defined in advance.

It aims to reduce the amount of information that needs to be built and managed for query analysis and to improve the accuracy of query analysis with relatively little data.

However, the technical problems to be achieved by the present embodiment are not limited to the technical problems described above, and other technical problems may exist.

As a means for achieving the above-described technical problem, an embodiment of the present invention provides a query analysis apparatus for analyzing query data, an input unit for receiving query data, and performing natural language processing on the received query data. A natural language processing unit for deriving 1 entity name pattern, a deep learning model processing unit for deriving a second entity name pattern using a deep learning model for the received query data, and for each entity name included in the second entity name pattern A similarity meta extraction unit for extracting similar metadata, a query analysis unit for performing query analysis on the received query data based on the first entity name pattern, the second entity name pattern, and the similar metadata, and the input It may include an output unit for outputting a query analysis result for the received query data.

In an embodiment, the received query data includes at least one keyword corresponding to a pronoun or a verb, and the natural language processing unit matches the keyword to any one of a plurality of entity names to match the received query data. It is possible to derive the first entity name pattern for .

In an embodiment, the natural language processing unit further includes a pattern accuracy deriving unit for deriving accuracy for each first entity name pattern when a plurality of first entity name patterns are derived, and the natural language processing unit is based on the accuracy Thus, one of the plurality of derived first entity name patterns may be selected.

In one embodiment, the deep learning model includes first learning data generated based on word embedding, second training data generated based on statistics on the first entity name pattern, and the query analysis result. It may be learned using one or more of the third learning data generated based on the feedback data.

In an embodiment, the deep learning model may include a first model that outputs a probability value corresponding to the received query data for a plurality of pre-learned entity name patterns.

In an embodiment, the deep learning model may further include a second model for deriving the second entity name pattern based on the output value of the first model.

In an embodiment, the similarity meta extraction unit may extract a word having a shape similar to each entity name included in the second entity name pattern as the similarity metadata based on the entity name index.

In an embodiment, when a word having a similar shape does not exist in the entity name index, the similarity meta extraction unit may extract a word having a similar shape as the similarity metadata based on word embedding.

Another embodiment of the present invention provides a query analysis method for analyzing query data, comprising: receiving query data; performing natural language processing on the received query data to derive a first entity name pattern; Deriving a second entity name pattern using a deep learning model with respect to the received query data, extracting similar metadata for each entity name included in the second entity name pattern, the first entity name pattern, The method may include performing a query analysis on the received query data based on the second entity name pattern and the similar metadata, and outputting a query analysis result on the received query data.

Another embodiment of the present invention provides a computer program stored in a medium including a sequence of instructions for analyzing query data, receiving query data and performing natural language processing on the received query data to name a first entity A pattern is derived, a second entity name pattern is derived using a deep learning model with respect to the received query data, similar metadata is extracted for each entity name included in the second entity name pattern, and the second entity name pattern is derived. A sequence of commands for performing query analysis on the received query data based on the first entity name pattern, the second entity name pattern, and the similar metadata and outputting the query analysis result on the received query data may include

The above-described problem solving means are merely exemplary, and should not be construed as limiting the present invention. In addition to the exemplary embodiments described above, there may be additional embodiments described in the drawings and detailed description.

According to any one of the above-described problem solving means of the present invention, query data including a plurality of keywords or query data including keywords not defined in advance can be analyzed.

It can also reduce the amount of information that must be built and managed for query analysis.

In addition, the accuracy of query analysis can be improved by using natural language processing and deep learning models.

In addition, a service can be provided using query analysis.

1 is a block diagram of a query analysis apparatus according to an embodiment of the present invention.
2 is a flowchart of a query analysis method according to an embodiment of the present invention.
3 is a flowchart of a method of deriving a first entity name pattern according to an embodiment of the present invention.
4 is a flowchart of a method of deriving a second entity name pattern according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily implement them. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a part is "connected" with another part, this includes not only the case of being "directly connected" but also the case of being "electrically connected" with another element interposed therebetween. . Also, when a part "includes" a component, it means that other components may be further included, rather than excluding other components, unless otherwise stated, and one or more other features However, it is to be understood that the existence or addition of numbers, steps, operations, components, parts, or combinations thereof is not precluded in advance.

In this specification, a "part" includes a unit realized by hardware, a unit realized by software, and a unit realized using both. In addition, one unit may be implemented using two or more hardware, and two or more units may be implemented by one hardware. Meanwhile, '~ unit' is not limited to software or hardware, and '~ unit' may be configured to be in an addressable storage medium or may be configured to reproduce one or more processors. Thus, as an example, '~' denotes components such as software components, object-oriented software components, class components, and task components, and processes, functions, properties, and procedures. , subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays and variables. The functions provided in the components and '~ units' may be combined into a smaller number of components and '~ units' or further separated into additional components and '~ units'. In addition, components and '~ units' may be implemented to play one or more CPUs in a device or secure multimedia card.

The "network" referred to below means a connection structure capable of exchanging information between each node, such as terminals and servers, and includes a local area network (LAN), a wide area network (WAN). , the Internet (WWW: World Wide Web), wired and wireless data networks, telephone networks, wired and wireless television networks, and the like. Examples of wireless data communication networks include 3G, 4G, 5G, 3rd Generation Partnership Project (3GPP), Long Term Evolution (LTE), World Interoperability for Microwave Access (WIMAX), Wi-Fi, Bluetooth communication, infrared communication, ultrasound Communication, Visible Light Communication (VLC), LiFi, etc. are included, but are not limited thereto.

Some of the operations or functions described as being performed by the terminal or device in the present specification may be instead performed by a server connected to the terminal or device. Similarly, some of the operations or functions described as being performed by the server may also be performed in a terminal or device connected to the server.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram of a query analysis apparatus according to an embodiment of the present invention. Referring to FIG. 1 , the query analysis apparatus 100 includes an input unit 110 , a natural language processing unit 120 , a deep learning model processing unit 130 , a similar meta extraction unit 140 , a query analysis unit 150 , and an output unit ( 160) may be included.

The query analysis apparatus 100 may analyze the query data and output a result value corresponding thereto. For example, the query analysis apparatus 100 may analyze query data for searching for content desired by the user in the media providing service and output a list of corresponding content as a result value.

The input unit 110 may receive query data. The query data may include text data and the like. The input unit 110 may receive, for example, natural language data as query data.

The query data may include at least one keyword. The keyword may correspond to, for example, a pronoun, such as a noun, a pronoun, or a verb, or a verb, such as a verb, an adjective, or the like. In the present invention, more accurate analysis results can be derived by analyzing query data including verbs as well as verbs generally included in keywords.

For example, when query data of "American movies about France" is input, the query data may include "France", "coming out", "USA", and "movies" as keywords. Here, "France", "United States" and "movie" are keywords corresponding to verbs, and "coming out" is keywords corresponding to verbs.

The natural language processing unit 120 may derive a first entity name pattern by performing natural language processing on the received query data. The natural language processing unit 120 may derive the first entity name pattern by using the entity name dictionary and the entity name pattern rule defined in the dictionary.

The natural language processing unit 120 may match the keyword to any one of a plurality of entity names. The natural language processing unit 120 may extract, for example, an entity name corresponding to a keyword from an entity name dictionary including information on a plurality of entity names. The entity name dictionary may include a plurality of predefined entity names such as a country, a content type, a spatial background, and a background attribute, and information about a plurality of words classified into each entity name.

The natural language processing unit 120 may analyze the first entity name pattern based on a predefined entity name pattern rule. The natural language processing unit 120 may use a plurality of predefined entity name pattern rules.

For example, for each keyword included in the query data of "American film with a French background", "France - Country", "Background - Background attribute", "US - Country", "Movie" based on the entity name dictionary. - You can match object names such as "content type". From this, it is possible to analyze the first entity name pattern such as "Spatial background: France, production country: USA, content type: movie" with reference to the entity name pattern rule defined in advance.

As another example, for each keyword included in the query data of "Lee Seung-gi's entertainment," based on the entity name dictionary, the entity name such as "Lee Seung-gi - person", "appearance - appearance attribute", "entertainment - content type" can be matched. From this, it is possible to analyze the first entity name pattern, such as "guest appearance: Seunggi Lee, content type: entertainment" or "caster: Seunggi Lee, content type: entertainment" with reference to the entity name pattern rule defined in advance.

Referring back to FIG. 1 , the natural language processing unit 120 may include a pattern accuracy deriving unit 121 . When a plurality of first entity name patterns are derived, the pattern accuracy deriving unit 121 may derive accuracy for each first entity name pattern.

The pattern accuracy deriving unit 121 may derive the accuracy of the first entity name pattern based on each preset weight.

For example, in the case of “Guest Appearance: Person, Content Type: Entertainment”, the weight is A1 × ( 1 / Last program date as a guest appearance) + B1 × (Type weight × Occupation weight) + C1 × (Guest appearance) number of times / (number of guest appearances + number of fixed appearances)), and in the case of “appearance: person, content type: entertainment,” the weight is A2 × ( 1 / date of last program appearance fixed) + B1 × It can be derived as (type weight × job weight) + C2 × (number of fixed appearances / (number of guest appearances + number of fixed appearances)). Here, the type weight is a value set by quantifying the probability that a person of each job will appear in the corresponding content, the job weight is a value set based on the job of each person, and A1, A2, B1, C1, and C2 are the results correction. It may be a value set for

The natural language processing unit 120 may select one of the plurality of first entity name patterns derived based on accuracy. The natural language processing unit 120 may, for example, select a first entity name pattern having a higher accuracy value from among the plurality of first entity name patterns.

The deep learning model processing unit 130 may derive the second entity name pattern by using the deep learning model with respect to the received query data.

The deep learning model processing unit 130 may train the deep learning model by using the first learning data, the second learning data, and the third learning data.

The training data of the deep learning model directly affects the performance of the deep learning model and the quality of the results. However, in most cases, it is difficult to construct training data. An object of the present invention is to provide a method for constructing learning data necessary for initial learning of a deep learning model and a method for continuously constructing learning data. In the case of a supervised learning-based deep learning model, learning data composed of pairs of problems and correct answers is required, so the present invention intends to automate the generation of learning data composed of pairs of problems and correct answers.

The deep learning model processing unit 130 may construct raw learning data by collecting query data from web documents such as web crawling and news. Such raw training data may not consist of pairs of questions and answers.

The deep learning model processing unit 130 may generate first training data based on word embedding. The deep learning model processing unit 130 may perform word embedding in which words and morphemes are transcribed into a multidimensional space with respect to raw learning data, externally collected data, and the like. The deep learning model processing unit 130 may use, for example, an algorithm such as Word2Vec, Phase2Vec, FastText, and may use it with some modifications.

The deep learning model processing unit 130 may generate the first learning data by comparing the word embedding result with the entity name dictionary. The deep learning model processing unit 130 may generate the first learning data further based on distant supervision.

For example, with respect to the word "classical", the word "classic" extracted by word embedding corresponds to the entity name "sentiment", and another extracted word "antique" corresponds to the entity name "material". In some cases, distant supervision may select a word that is more similar in form, "old-fashioned". Through this process, the first learning data in which the word "classic" is matched with the name of the entity "sentiment" is generated.

The deep learning model processing unit 130 may generate second learning data based on statistics on the first entity name pattern.

The deep learning model processing unit 130 may use, for example, a first entity name pattern derived from a query that the user actually inputs a lot.

The deep learning model processing unit 130 may generate the second learning data, for example, based on statistics on the first entity name pattern and the entity name dictionary. For example, the second learning data may be generated by combining words corresponding to each entity name with respect to an entity name pattern including a material, an appearance attribute, a genre, and a content type.

The deep learning model processing unit 130 may generate third learning data based on the feedback data for the query analysis result. The deep learning model processing unit 130 may generate third learning data, for example, based on feedback data input from the manager of the query analysis apparatus 100 . Reliability and accuracy of the query analysis result may be improved by the third learning data.

The deep learning model processing unit 130 may use a deep learning model including a first model and a second model. The first model may be, for example, based on BiLSTM-CRF.

The deep learning model processing unit 130 may output a probability value corresponding to the received query data for a plurality of pre-learned entity name patterns using the first model. For example, when the deep learning model processing unit 130 receives "action movie in which a detective appears" as query data using the first model, the entity name pattern "Material: detective, appearance property: appearing, genre: action , content type: movie”, and another entity name pattern “character: detective, appearance attribute: coming out, genre: action, content type: movie” with a probability of 20%. can be printed out.

The deep learning model processing unit 130 may derive the second entity name pattern by using the second model. The second model may be one that has been trained to output an entity name pattern corresponding to the correct answer by taking the output value of the first model as an input.

The deep learning model processing unit 130 may serve to extract an entity name pattern that can be actually input as a user's query from among entity name patterns output from the first model by using the second model. The deep learning model processing unit 130 may compensate for the case in which the first model is not sufficiently trained by using the second model.

The similarity meta extraction unit 140 may extract similar metadata for each entity name included in the second entity name pattern. The similarity meta extraction unit 140 may extract a noun, an adjective, etc. determined to be important morphemes from the entity name dictionary and use the entity name index indexed by entity name type. The similarity meta extractor 140 may index the keyword “glamorous” as, for example, 'Key-glamorous, Value-glamorous'. The similarity meta extractor 140 may index, for example, 'Key- creepy, Key- creepy, Value- creepy'' with respect to the keyword "creepy". The similarity meta extraction unit 140 may extract a noun, an adjective, etc. determined as important morphemes from the second entity name pattern, and compare it with an entity name index.

The similarity meta extraction unit 140 may extract a word having a shape similar to each entity name included in the second entity name pattern as similarity metadata based on the entity name index.

When a word having a similar shape does not exist in the entity name index, the similarity meta extraction unit 140 may extract a word having a similar shape as similar metadata based on word embedding.

The query analysis unit 150 may perform query analysis on the received query data based on the first entity name pattern, the second entity name pattern, and similar metadata.

The output unit 160 may output a query analysis result for the received query data. The output unit 160 may output a query analysis result using, for example, a speaker or a display.

In an embodiment, the query analysis apparatus 100 may further include a determination unit (not shown). The determination unit may determine whether to directly output the first entity name pattern without deriving the second entity name pattern with respect to the received query data. For example, as a result of determining the derived first entity name pattern, the determination unit may determine to derive the second entity name pattern for the received query data without outputting the first entity name pattern.

2 is a flowchart of a query analysis method according to an embodiment of the present invention. The method 200 for analyzing a query performed by the query analyzing apparatus 100 shown in FIG. 2 includes steps that are time-series processed by the query analyzing apparatus 100 according to the embodiment shown in FIG. 1 . Therefore, even if omitted below, it is also applied to the method of analyzing a query performed by the query analyzing apparatus 100 according to the embodiment shown in FIG. 1 .

In step S210, the query analysis apparatus 100 may receive query data.

In step S220, the query analysis apparatus 100 may derive a first entity name pattern by performing natural language processing on the received query data.

In step S230 , the query analysis apparatus 100 may derive a second entity name pattern from the received query data using a deep learning model.

In operation S240, the query analyzing apparatus 100 may extract similar metadata for each entity name included in the second entity name pattern.

In operation S250, the query analysis apparatus 100 may perform query analysis on the received query data based on the first entity name pattern, the second entity name pattern, and similar metadata.

In operation S260 , the query analysis apparatus 100 may output a query analysis result for the received query data.

In the above description, steps S210 to S260 may be further divided into additional steps or combined into fewer steps, according to an embodiment of the present invention. In addition, some steps may be omitted if necessary, and the order between the steps may be switched.

3 is a flowchart of a method of deriving a first entity name pattern according to an embodiment of the present invention. The method 300 for deriving the first entity name pattern performed by the query analyzing apparatus 100 shown in FIG. 3 is time-series processing by the query analyzing apparatus 100 according to the embodiment shown in FIG. 1 . include those Therefore, even if omitted below, it is also applied to the method of deriving the first entity name pattern performed by the query analyzing apparatus 100 according to the embodiment shown in FIG. 1 .

In step S310, the query analysis apparatus 100 may receive query data.

In operation S320 , the query analyzing apparatus 100 may derive a plurality of first entity name patterns for the received query data.

In step S330 , the query analyzing apparatus 100 may derive accuracy for each first entity name pattern.

In operation S340 , the query analyzing apparatus 100 may select one of the plurality of first entity name patterns based on accuracy.

In the above description, steps S310 to S340 may be further divided into additional steps or combined into fewer steps, according to an embodiment of the present invention. In addition, some steps may be omitted if necessary, and the order between the steps may be switched.

4 is a flowchart of a method of deriving a second entity name pattern according to an embodiment of the present invention. A flowchart 400 of a method of deriving a second entity name pattern performed by the query analyzing apparatus 100 shown in FIG. 4 is time-series processed by the query analyzing apparatus 100 according to the embodiment shown in FIG. 1 . includes the steps to be Therefore, even if omitted below, the flowchart of the method for deriving the second entity name pattern performed by the query analysis apparatus 100 according to the embodiment shown in FIG. 1 is also applied.

In step S410, the query analysis apparatus 100 may receive query data.

In step S420 , the query analysis apparatus 100 may derive a first entity name pattern by performing natural language processing on the received query data.

In step S430 , the query analysis apparatus 100 may derive a second entity name pattern from the received query data using a deep learning model.

In the above description, steps S410 to S430 may be further divided into additional steps or combined into fewer steps according to an embodiment of the present invention. In addition, some steps may be omitted if necessary, and the order between the steps may be switched.

The method of analyzing query data in the query analyzing apparatus described with reference to FIGS. 1 to 4 may be implemented in the form of a computer program stored in a medium executed by a computer or a recording medium including instructions executable by the computer. . Also, the method of analyzing query data in the query analyzing apparatus described with reference to FIGS. 1 to 4 may be implemented in the form of a computer program stored in a medium executed by a computer.

Computer-readable media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. Also, computer-readable media may include computer storage media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data.

The above description of the present invention is for illustration, and those of ordinary skill in the art to which the present invention pertains can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and likewise components described as distributed may be implemented in a combined form.

The scope of the present invention is indicated by the following claims rather than the above detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention. do.

100: query analysis device
110: input unit
120: natural language processing unit
130: deep learning model processing unit
140: similar meta extraction unit
150: query analysis unit
160: output unit

Claims (17)

In the query analysis device for analyzing query data,
an input unit for receiving query data;
a natural language processing unit for deriving a first entity name pattern by performing natural language processing on the received query data;
a deep learning model processing unit for deriving a second entity name pattern using a deep learning model with respect to the received query data;
a similarity meta extraction unit for extracting similar meta data for each entity name included in the second entity name pattern;
a query analysis unit that performs query analysis on the received query data based on the first entity name pattern, the second entity name pattern, and the similar metadata; and
An output unit for outputting a query analysis result for the received query data
That comprising a, query analysis device.
The method of claim 1,
The received query data includes at least one keyword corresponding to a verb or a verb,
and the natural language processing unit derives the first entity name pattern for the received query data by matching the keyword to any one entity name among a plurality of entity names.
The method of claim 1,
The natural language processing unit is a pattern accuracy deriving unit that derives accuracy for each first entity name pattern when a plurality of first entity name patterns are derived.
further comprising,
wherein the natural language processing unit selects one of the plurality of derived first entity name patterns based on the accuracy.
The method of claim 1,
The deep learning model is based on first training data generated based on word embedding, second training data generated based on statistics on the first entity name pattern, and feedback data for the query analysis result. A query analysis device that is learned using one or more of the generated third learning data.
5. The method of claim 4,
The deep learning model includes a first model that outputs a probability value corresponding to the received query data with respect to a plurality of pre-learned entity name patterns.
6. The method of claim 5,
The deep learning model further comprises a second model for deriving the second entity name pattern based on the output value of the first model, query analysis apparatus.
The method of claim 1,
and the similarity meta extracting unit extracts, as the similarity metadata, a word having a shape similar to each entity name included in the second entity name pattern, based on the entity name index.
8. The method of claim 7,
The similarity meta extraction unit extracts, as the similarity metadata, a word having a similar shape based on word embedding when there is no word having a similar shape in the entity name index.
In the query analysis method of analyzing query data,
receiving query data;
deriving a first entity name pattern by performing natural language processing on the received query data;
deriving a second entity name pattern using a deep learning model with respect to the received query data;
extracting similar metadata for each entity name included in the second entity name pattern;
performing query analysis on the received query data based on the first entity name pattern, the second entity name pattern, and the similar metadata; and
outputting a query analysis result for the received query data
A query analysis method comprising a.
10. The method of claim 9,
The received query data includes at least one keyword corresponding to a verb or a verb,
The step of deriving the first entity name pattern comprises deriving the first entity name pattern for the received query data by matching the keyword to any one entity name among a plurality of entity names.
10. The method of claim 9,
deriving accuracy for each first entity name pattern when a plurality of first entity name patterns are derived; and
selecting one of the derived plurality of first entity name patterns based on the accuracy
Further comprising a, query analysis method.
10. The method of claim 9,
The deep learning model is based on first training data generated based on word embedding, second training data generated based on statistics on the first entity name pattern, and feedback data for the query analysis result. A query analysis method that is learned using one or more of the generated third learning data.
13. The method of claim 12,
The deep learning model includes a first model that outputs a probability value corresponding to the received query data with respect to a plurality of pre-learned entity name patterns.
14. The method of claim 13,
The deep learning model further comprises a second model for deriving the second entity name pattern based on the output value of the first model.
10. The method of claim 9,
The extracting of the similarity metadata comprises extracting, as the similarity metadata, a word having a shape similar to each entity name included in the second entity name pattern based on the entity name index.
16. The method of claim 15,
The extracting of the similarity metadata includes extracting a word having a similar shape as the similarity metadata based on word embedding when a word having a similar shape does not exist in the entity name index.
A computer program stored on a medium comprising a sequence of instructions for analyzing query data, the computer program comprising:
When the computer program is executed by a computing device,
Receive query data,
A first entity name pattern is derived by performing natural language processing on the received query data,
Deriving a second entity name pattern using a deep learning model with respect to the received query data,
extracting similar metadata for each entity name included in the second entity name pattern;
performing query analysis on the received query data based on the first entity name pattern, the second entity name pattern, and the similar metadata;
A computer program stored in a medium comprising a sequence of instructions for outputting a query analysis result for the received query data.

Images