KR20220037059A - A method and apparatus for question-answering using multiple similarity determination for query vectors - Google Patents

Abstract

The present invention discloses a question-answering method performed by an electronic apparatus. The method comprises: a step of generating a user query vector for a user query; a step of transmitting the user query vector to a server; a step of receiving similar query candidates selected based on a similarity with the user query vector from the server; a step of selecting a similar query from the similar query candidates based on an analysis result of the similarity with the user query vector; and a step of determining an answer for the similar query as an answer for the user query.

Description

A search method and apparatus for question-answering using multiple similarity determination for query vectors

The present invention relates to a method of retrieving a response to a user's query and presenting it to a user.

Search engines are being used to search for information using the Internet. As a method for a search engine to provide an answer to a user's query, a method of providing a database search result based on text information constituting a user query input by a user is widely used.

However, in the method of providing search results for a user query based on text information constituting the user query, if the expression method of the query in the query-response pair stored in the database is different from the query input by the user, the search engine returns the search result There is a problem that it cannot be derived with .

In order to solve this problem, semantic labeling is performed by performing linguistic analysis on text documents or user questions in Korean Patent Application Laid-Open No. 10-2011-0134909, etc., and to find texts related to semantically labeled user questions. Methods for searching a labeled text document database are presented.

However, the above method has a problem in that a large amount of text document database must be searched to find a response corresponding to a user request, and the search time increases as the size of the text document database increases.

In this specification, it is intended to propose a method of efficiently searching for a response to a user's query through multiple similarity determinations with respect to a query vector.

In order to solve the above problem, a search method performed by an electronic device according to an embodiment of the present invention includes generating a user query vector for a user query; transmitting the user query vector to a server; receiving, from the server, a similar query candidate selected based on the similarity with the user query vector; selecting a similar query from among the similar query candidates based on a similarity analysis result with the user query vector; and determining a response to the similar query as a response to the user query.

The similar query candidate may be determined based on a similarity between a query vector stored in the server and a user query vector transmitted from the electronic device to the server.

The similarity query candidate is selected by the server based on similarity with the user query vector by a first similarity analysis model, and the similarity query is based on a similarity analysis result with the user query vector by a second similarity analysis model to be selected by the electronic device, and the first similarity analysis model and the second similarity analysis model may be different analysis models.

Selecting a similar query from among the similar query candidates based on the similarity analysis result with the user query vector includes determining the similarity between the vector of the similar query candidate and the user query vector using a similarity analysis model. and the similarity analysis model may be trained using a Siamese neural network.

The similarity analysis model is trained using first training data and second training data, wherein the first training data includes a first query pair and a label indicating that the first query pair is similar, and the second training data may be composed of a label indicating that the second query pair is dissimilar to the second query pair.

Receiving the similar query candidate selected based on the degree of similarity with the user query vector from the server may include receiving a query response corresponding to the similar query candidate.

Receiving the similar query candidate selected based on the similarity with the user query vector from the server may include receiving a similarity between the vector of the similar query candidate and the user query vector.

The selecting of a similar query from among the similar query candidates based on the similarity analysis result with the user query vector may include determining a similarity ranking of candidate queries belonging to the similar query candidate based on the similarity analysis result with the user query vector. to do; and selecting a predetermined number of candidate queries as similar queries according to the similarity ranking.

Receiving the similar query candidate selected based on the degree of similarity with the user query vector from the server may include receiving a query response corresponding to the similar query candidate.

A search method performed by an electronic device according to an embodiment of the present invention includes: transmitting the selected similar query to the server; and receiving a response to the similar query from the server.

In addition, a search apparatus according to an embodiment of the present invention for solving the above problem includes: a processor; and a memory for storing instructions to be executed by the processor, wherein the processor generates a user query vector for a user query by executing the instruction, transmits the user query vector to a server, and the user A similar query candidate selected based on the similarity with the query vector is received from the server, a similar query is selected from among the similar query candidates based on the similarity analysis result with the user query vector, and the response to the similar query is received from the server. It can be decided in response to user queries.

In addition, a computer program stored in a computer-readable storage medium according to an embodiment of the present invention for solving the above problem includes the steps of: generating a user query vector for a user query; transmitting the user query vector to a server; receiving, from the server, a similar query candidate selected based on the similarity with the user query vector; selecting a similar query from among the similar query candidates based on a similarity analysis result with the user query vector; and determining a response to the similar query as a response to the user query.

The present invention provides an effect of presenting a search result for a query similar to the user's query to the user. Furthermore, the present invention provides an effect that similar queries can be efficiently selected by the search server first selecting similar queries and the search terminal secondarily verifying the similarity of similar queries.

1 is a block diagram illustrating a search system according to an embodiment of the present invention.
2 and 3 are block diagrams illustrating a terminal and a server according to an embodiment of the present invention.
4 is a flowchart illustrating a search method according to an embodiment of the present invention.
5 and 6 are diagrams illustrating an example of a process of generating an analysis model by learning a neural network.
7 is a diagram illustrating an example of generating a vector by embedding a user query expressed in natural language.
8 shows a neural network learning process for generating an encoder.
9 is a diagram for explaining the concept of a Siamese neural network.
10 is a flowchart illustrating a method for generating a paraphrase model by an analysis model generating apparatus according to an embodiment of the present invention.
11 is a diagram for explaining a vector-based search method of a server.

The following is merely illustrative of the principles of the invention. Therefore, those skilled in the art can devise various devices that, although not explicitly described or shown herein, embody the principles of the invention and are included in the spirit and scope of the invention. In addition, it should be understood that all conditional terms and examples listed herein are, in principle, expressly intended only for the purpose of understanding the inventive concept and are not limited to the specifically enumerated embodiments and states as such. .

The above-described objects, features and advantages will become more apparent through the following detailed description in relation to the accompanying drawings, and accordingly, those of ordinary skill in the art to which the invention pertains will be able to easily practice the technical idea of the invention. .

In the specification and claims, terms such as "first", "second", "third", and "fourth" are used, if any, to distinguish between like elements, and are not necessarily used in a specific sequence, if any. or to describe the order of occurrence. It will be understood that the terms so used are interchangeable under appropriate circumstances to enable the embodiments of the invention described herein to operate, for example, in sequences other than those shown or described herein. Likewise, where methods are described herein as including a series of steps, the order of those steps presented herein is not necessarily the order in which those steps may be performed, and any described steps may be omitted and/or Any other steps not described may be added to the method.

Also, terms such as "left", "right", "front", "behind", "top", "bottom", "above", "below" in the specification and claims are used for descriptive purposes, It is not necessarily intended to describe an invariant relative position. It will be understood that the terms so used are interchangeable under appropriate circumstances to enable the embodiments of the invention described herein to operate otherwise than, for example, as shown or described herein. As used herein, the term “connected” is defined as being directly or indirectly connected in an electrical or non-electrical manner. Objects described herein as being "adjacent" to one another may be in physical contact with one another, in proximity to one another, or in the same general scope or area as appropriate for the context in which the phrase is used. The presence of the phrase “in one embodiment” herein refers to the same, but not necessarily, embodiment.

In addition, in the specification and claims, references to 'connected', 'connecting', 'fastened', 'fastening', 'coupled', 'coupled', etc., and various variations of these expressions, refer to other elements directly It is used in the sense of being connected or indirectly connected through other elements.

In addition, the suffixes "module" and "part" for components used in this specification are given or used in consideration of ease of writing the specification, and do not have a meaning or role distinct from each other by themselves.

In addition, the terms used herein are for the purpose of describing the embodiments and are not intended to limit the present invention. In this specification, the singular also includes the plural unless specifically stated otherwise in the phrase. As used herein, 'comprise' and/or 'comprising' means that a referenced component, step, operation and/or element is the presence of one or more other components, steps, operations and/or elements. or addition is not excluded.

In addition, in the description of the invention, if it is determined that a detailed description of a known technology related to the invention may unnecessarily obscure the gist of the invention, the detailed description thereof will be omitted. Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram illustrating a search system according to an embodiment of the present invention. The terminal 100 according to an embodiment transmits a query to the server 200 and receives a response to the query from the server 200 . The terminal 100 provides an answer to the user's query based on the response to the query received from the server 200 .

The server 200 may be a search engine that performs elastic search based on keywords. In an embodiment, the server 200 may search for a similar query based on a vector. Through this, the server 200 may select a query having a high degree of similarity to the user query and provide it to the terminal 100 . The server 200 primarily selects a similar query candidate group for a user query, and the terminal 100 performs a similarity evaluation on a query belonging to the similar query candidate group provided by the server 200, so that the terminal 100 is the user In order to select a similar query similar to a query, the number of queries to be analyzed for similarity with a user query may be limited to a certain range.

2 is a block diagram illustrating the terminal 100 according to an embodiment of the present invention. The terminal 100 according to an embodiment of the present invention may include an encoder 110 , a parafraser 120 , a similar query selecting unit 130 , a query answering unit 140 , and a database 150 .

The encoder 110 generates an embedding vector corresponding to a user query expressed in natural language. An embedding vector is a vector representing the structure and meaning of a user query expressed in natural language. The encoder 110 may be configured as a neural network, and may be a neural network model trained to generate an embedding vector.

The Parafraser 120 is a similarity analysis model that analyzes the similarity of a plurality of queries expressed in natural language. The Parafraser 120 receives two natural language queries as inputs, and provides similarity values of the queries as outputs. For example, the Parafraser 120 may vectorize two queries, and determine a similarity value based on a distance or cosine similarity of vectors for each query in Euclidean space. The parafraser 120 may be configured to include the encoder 110 to vectorize a query expressed in natural language. The Parafraser 120 may be configured as a neural network, and may be a neural network model trained to determine the similarity between queries.

The similar query selector 130 selects a query similar to the user's query according to the degree of similarity obtained by the parafraser 120 . And the similarity level of the similar query is determined according to the order similar to the user's query.

The query answering unit 140 obtains a user query from the user. In addition, the query answering unit 140 provides an answer to the user's query to the user. The query answering unit 140 provides the user with an answer to the inputted user's query when the inputted user's query and its answer are stored in the database 150 . On the other hand, when the input user query is not stored in the database 150, the query answering unit 140 corresponds to the similarity level according to a preset criterion according to the similarity level selected by the similar query selecting unit 130. An answer to a similar query may be provided to the user as the user's answer.

On the other hand, the terminal 100 and the server 200 according to an embodiment of the present invention, as shown in FIG. 3 , the processor 101 , the memory 102 , the communication unit 103 , the user interface 104 , and the database 105 . It may consist of including. A program for performing the operation of the search method according to an embodiment of the present invention may be stored in the memory 102 . A neural network model trained to implement the encoder 110 and the parafraser 120 may be stored in the memory 102 . The processor 101 controls the communication unit 103, the user interface 104, and the database 105 by processing data according to the program and the neural network model stored in the memory 102, thereby performing a search method according to an embodiment of the present invention. can be done

Hereinafter, a search method according to an embodiment of the present invention will be described. 4 is a flowchart illustrating a search method according to an embodiment of the present invention. The terminal 100 obtains a user's query based on the user input (S110). The terminal 100 may obtain a user query through a user's text input input through a keyboard or a touch pad or a voice input by the user's utterance. The user query may be a sentence expressed in a natural language.

Next, the terminal 100 vectorizes the user query expressed in natural language (S120). The terminal 100 may generate a user query vector by converting the user query into a vector representation using the encoder 110 . Next, the terminal 100 transmits the user query vector to the server 200 (S130).

The server 200 searches for a similar query using the user query vector received from the terminal 100 (S140). The server 200 searches for a query having the same or similar vector as the user query vector from among the queries stored in the database of the server 200 . In more detail, the database of the server 200 may store a query and a response to the query in association. For example, (question, response) pairs may be stored in the database. In addition, the database may store a query and a vector for the query in association. For example, (question, response, query vector) pairs may be stored in the database.

In order to search for a query having a similar vector, the server 200 may determine a vector located within a predetermined distance from the user query vector in the Euclidean space as a similar vector. Alternatively, the server 200 may determine the similarity based on the cosine similarity. The server 200 generates a similar query candidate group by using the similar query search result. The similar query candidate group may consist of only a query or may include at least one of a query, a response, and a vector. Also, the similarity query candidate group may further include a similarity value with the user query vector for each query belonging to the candidate group. Furthermore, the similarity query candidate group may further include an order of similarity between queries belonging to the candidate group. The server 200 transmits the generated similar query candidate group to the terminal 100 (S150).

Next, the terminal 100 selects a similar query by using the similar query candidate group information received from the server 200 ( S160 ). The terminal 100 may select a query similar to the user query from among the queries belonging to the similar query candidate group. The terminal 100 may determine the similarity between each query belonging to the similar query candidate group and the input user query by using the parafraser 120 . The terminal 100 may select a query having the highest similarity as a similar query. Alternatively, the terminal 100 may select a similar query for each grade by grading the queries belonging to the similar query candidate group in the order of the highest similarity. The similarity analysis method by the parafraser 120 may be different from the similarity analysis method performed by the server 200 . For example, the similarity analysis model of the Parafraser 120 may be a different analysis model from the similarity analysis model of the server 200, and in the case of a neural network model, the values of variables constituting the model configuration or the hidden layer are different from each other. can be different.

Next, the terminal 100 provides a response according to the selected similar query (S170). The terminal 100 may provide a response corresponding to a similar query having the highest similarity as a response to a user query input by the user. Alternatively, the terminal 100 may provide a response corresponding to a query belonging to a similar query candidate group as a response to a user query input by the user according to the similar query level.

The terminal 100 may not receive all responses to the similar query candidate group from the server 200 , but only the responses to the selected similar query from the server 200 . In this case, response information is not included in the similar query candidate group, and the terminal 100 transmits a similar query requiring response information to the server 200 and receives a response to the similar query from the server, thereby only responding to the similar query. It may be implemented by receiving transmission from the server 200 . Accordingly, traffic between the terminal 100 and the server 200 may be reduced.

A similar query may be a query that has a different meaning from a user query. For example, when the user inputs a query “How long is it to the city hall?”, the terminal 100 responds to similar queries such as 'How long does it take to reach the city hall' and 'What is the distance to get to the city hall? ' Or, you can select a similar query, such as 'How much is the transportation cost to get to City Hall?' In addition, in the case of similar queries such as 'How long does it take to get to City Hall?', 'How much is the transportation cost to reach City Hall by car?' and 'How much is the transportation cost to reach City Hall by public transportation?' Or, a similar query may exist, such as 'how much does it cost to travel to city hall on foot?'

In this case, when only answers to similar queries are provided to the user, the user may misunderstand a response with a different meaning as an answer to a query input by the user. Accordingly, when the terminal 100 provides the user with a response corresponding to the similar query, the terminal 100 may provide the user with the response to the similar query together with the similar query.

Meanwhile, in the description with reference to FIG. 4 , step S130 may be performed by the terminal 100 transmitting the user query itself, not the user query vector, to the server 200 . In this case, the server 200 may generate a user query vector for the user query using the encoder. Then, in step S140, the server 200 may search for a similar query by using the generated user query vector.

Hereinafter, each step will be described in more detail with reference to the drawings. The terminal 100 and the server 200 according to an embodiment of the present invention may vectorize a query by analyzing a natural language, and store the trained neural network model to analyze the similarity between the queries in a vector dimension. The neural network model according to an embodiment may be generated at a scale capable of operating according to an application field or computing power of a device.

In an embodiment, the neural network model may be a deep learning model. A deep learning model can be built with multiple hidden layers. Nodes belonging to a layer can send and receive data according to the connection relationship between nodes. The terminal 100 and the server 200 according to an embodiment of the present invention generate an analysis model by learning a deep neural network (DNN), a recurrent neural network (RNN), a bidirectional recurrent deep neural network (BRDNN), etc. The method according to an embodiment may be performed using the analyzed model.

5 is a flowchart illustrating a process of generating an analysis model by learning a neural network. The encoder 110 and the parafraser 120 according to an embodiment may be generated by the analysis model generating apparatus according to the analysis model generating method illustrated in FIG. 5 . First, the analysis model generating apparatus acquires learning data (S210). The training data may be set according to the analysis model to be generated.

Next, the analysis model generating apparatus trains the neural network using the acquired learning data (S220). It will be described in more detail with reference to FIG. 6 . The apparatus for generating an analysis model may set a neural network model according to an analysis model to be generated, and may determine a predicted value by performing forward propagation. In addition, the apparatus for generating an analysis model may determine an error value for determining a value of a variable by comparing a predicted value with an actual value using a loss function. The device may adjust the variable values of the neural network model using backward propagation so that the error between the predicted value and the actual value is reduced. As an optimizer used for backpropagation, a batch gradient descent method, a stochastic gradient descent method, or the like may be used.

After the neural network is trained, the apparatus for generating an analysis model acquires test data to check whether the neural network derives a valid analysis result (S230), and tests the neural network using the acquired test data (S240). The test data may consist of input values and corresponding actual values. The apparatus for generating an analysis model terminates learning of the neural network when the predicted value, which is a result of applying the input value of the test data to the neural network, has an error value within the allowable error range from the actual value of the test data, the learned neural network can be used as an analysis model. (S250). Meanwhile, the apparatus for generating an analysis model may continue to learn the neural network when the error value between the predicted value and the actual value according to the input of the test data is out of the allowable error range.

Meanwhile, the analysis model generating apparatus may apply early termination or dropout in order to prevent overfitting. In an embodiment, the apparatus for generating an analysis model trains an epoch before the accuracy is lowered if it is found that the accuracy of the verification data is continuously lowered while monitoring the error value between the predicted value and the actual value of the verification data. Early termination can be implemented by setting it as an epoch. Alternatively, the device may prevent overfitting by performing dropout of nodes that are not used for learning based on a certain criterion. In this way, the device can prevent the neural network from changing dependently on a specific node or combination of nodes.

Hereinafter, a vectorization method of a query performed by the encoder 110 will be described. The encoder 110 may generate a vector by embedding a user query expressed in natural language as shown in FIG. 7 . The encoder 110 according to an embodiment of the present invention is a vector generation model generated using a neural network, and may receive a query as an input and output a vector value corresponding to the query. The encoder 110 according to an embodiment may be a model such as Bag of Words (BOW), TF-IDF, Doc2Vec, Text Embedding Encoder, or Word2Vec.

8 shows a neural network learning process for generating the encoder 110 . The analysis model generation device trains a neural network model such as a DNN or RNN model using the preprocessed training data (S310) (S320), and tests the trained model (S330, S340) to generate the encoder 110 model. There is (S350).

For example, the analysis model generating apparatus may pre-process the training data before generating the encoder 110 . The analysis model generating apparatus may extract words by tokenizing the query. In an embodiment, the analysis model generating apparatus may perform tokenization based on a space, a colon, a comma, or the like. Additionally, the analysis model generating apparatus may perform preprocessing through morpheme analysis, removal of stopwords, and the like.

Next, in an embodiment, the apparatus for generating an analysis model may learn a Word2Vec model by applying a continuous bag of words (CBOW). The analysis model generating device applies the score vector according to the one-hot vector of the surrounding word and the one-hot vector of the central word to the cross-entropy function, which is a loss function, and learns so that the value of the loss function approaches 0. The encoder 110 model can be generated by modifying the parameters of the neural network by proceeding with the process.

Hereinafter, a method of determining the similarity of a query pair performed by the parafraser 120 will be described. The paraphrase 120 is a paraphrase model that outputs the similarity of the query pair with respect to the input of the query pair. The analysis model generating apparatus may build a paraphrase model by learning the neural network model. The neural network model may be CNN, DNN, RNN, or the like.

The paraphrase model according to an embodiment may acquire a vector expression of a query using the encoder 110 described above. The paraphrase model can evaluate the similarity between two queries by evaluating the vector similarity. For example, the paraphrase model can evaluate the similarity between two queries by generating a vector representation of the query and calculating the distance between the two query vectors in the Euclidean space. The paraphrase model can determine that two query vectors are similar as far as they are close.

In another embodiment, the paraphrase model may evaluate the similarity by calculating the cosine similarity between the vector representations of two queries. The paraphrase model may determine that the value of the cosine similarity is closer to 1, and the closer to -1, the more dissimilar.

Meanwhile, in the case of analyzing the similarity of the queries in the above manner, the queries may be analyzed to be similar to each other even if the queries are similar in form but have completely different contents by changing some words. For example, the query 'How much is the transportation cost from here to the city hall?' and the query 'What is the distance from here to the city hall?' The similarity analysis method described above may determine that the query is similar.

As such, the parafraser 120 must be able to identify that the queries included in the query are similar to each other but have semantically different contents that are dissimilar to each other. The paraphrase model according to an embodiment uses a siamnese neural network and may be learned by supervised learning.

9 is a diagram for explaining the concept of a Siamese neural network. A Siamese neural network consists of the same neural network. The analysis model generating device changes the variables constituting the Siamese neural network so that, when a query with the same meaning is input, similar vectors are generated as result values, and when queries with different meanings are input, dissimilar vectors are generated as result values. You can create a paraphrase model by training the neural network as you go. Thus, the parafraser 120 may determine the similarity of both queries by analyzing the similarity of vector values generated for each query. Here, the vector similarity may be determined based on how close the value of the distance proximity or cosine similarity in the Euclidean space is to 1.

10 is a flowchart illustrating a method for generating a paraphrase model by an analysis model generating apparatus according to an embodiment of the present invention. In the following description, it is described that the query constituting the learning data is expressed as text, but according to an embodiment, the query constituting the learning data may be expressed as a vector by the encoder described above.

The apparatus for generating an analysis model according to an embodiment of the present invention generates first learning data composed of similar words and structures so that sentences have a similar form and also have similar meanings of queries ( 410 ). The first training data may also include a label indicating that the corresponding queries have similar meanings along with the query pair. For example, training data for training a paraphrase model may be composed of a query pair and a similarity label of the query pair. For example, the training data may be composed of a query pair composed of (first query, second query) and a similarity score of the query pair.

Next, the analysis model generating apparatus performs learning of the neural network using the first learning data (S420). The apparatus for generating an analysis model may check the degree of learning by using the test data as the learning progress progresses, and may stop learning when overfitting starts to occur.

Next, the analysis model generating apparatus generates second learning data composed of queries that are composed of similar words and structures and have similar sentence forms, but have different meanings of the queries ( 430 ). The second training data may also include a label indicating that the queries do not have similar meanings along with the query pair.

Next, the analysis model generating apparatus performs learning of the neural network using the second learning data (S440). The apparatus for generating an analysis model may check the degree of learning by using the test data as the learning progress progresses, and may stop learning when overfitting starts to occur.

Meanwhile, neural network learning in steps S420 and S440 is performed with learning data having different labels. The apparatus for generating an analysis model may train the neural network without distinction between steps S420 and S440 by mixing the first learning data and the second learning data in order to increase learning efficiency. The analysis model generating apparatus may use a triplet loss function or the like to determine an error value between the predicted value and the result value.

Next, the analysis model generating device inputs the test data to the neural network trained as above. When generating a predicted value within an error range with respect to the test data, the analysis model generating apparatus may determine that the learned neural network can be used as a paraphrase model (S450).

Hereinafter, a vector-based search method of the server 200 will be described. The server 200 according to an embodiment of the present invention may store a query response pair consisting of a query and a response in the database, and a query vector, which is a vector expression for a query, may also be stored in the database together with the query and response. there is. For example, a pair of (question, response, and query vector) may be stored in the database of the server 200 . The encoder described above can be used to generate the query vector. The server 200 may generate a query vector for a query using the encoder described above.

A vector-based search method of the server 200 will be described in more detail with reference to FIG. 11 . The server 200 may pre-process the query (S510). In an embodiment, all uppercase letters may be converted to lowercase letters. Stopwords can be removed. For example, stopwords such as a, an, and the can be removed. In addition, a morpheme analysis process may be performed.

Next, the server 200 vectorizes the query (S520). For vectorization, the previously described encoder 110 model may be used. The device may use the encoder 110 model to generate the vector value of the query.

Next, the server 200 adds the vector to the database (S530). Accordingly, the database may include queries, responses, and vector values for queries.

Accordingly, the server 200 may perform a query search based on the query vector (S540). For example, the server 200 may search for the same query based on the text of the query input by the user, or may present a query including a similar keyword based on the text as a search result.

In addition, the server 200 may vectorize a query input by the user, search a database based on the generated vector, and present a query having the same or similar vector value to the user. The similarity of vectors may be determined as a distance or cosine similarity in Euclidean space.

Meanwhile, the server 200 may receive a vector value from the terminal 100 . In this case, the server 200 may search the database based on the input vector value and provide a pair of a query having the same or similar vector value and a response thereof to the terminal 100 .

The search method according to an embodiment described above may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured according to the embodiment, or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic such as floppy disks. - a hardware platform diet management server 10 specially configured to store and execute magneto-optical media, and program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.

Each of the drawings referenced in the description of the previous embodiment is merely an embodiment shown for convenience of description, and items, contents, and images of information displayed on each screen may be modified and displayed in various forms.

Although the present invention has been described with reference to an embodiment shown in the drawings, this is merely exemplary, and those of ordinary skill in the art will understand that various modifications and equivalent other embodiments are possible therefrom. Accordingly, the true technical protection scope of the present invention should be determined by the technical spirit of the appended claims.

Claims (12)

A search method performed by an electronic device, comprising:
generating a user query vector for the user query;
transmitting the user query vector to a server;
receiving, from the server, a similar query candidate selected based on the similarity with the user query vector;
selecting a similar query from among the similar query candidates based on a similarity analysis result with the user query vector; and
and determining a response to the similar query as a response to the user query.
The method of claim 1,
The similar query candidate is determined based on a similarity between a query vector stored in the server and a user query vector transmitted from the electronic device to the server.
The method of claim 1,
The similarity query candidate is selected by the server based on the similarity with the user query vector by a first similarity analysis model;
The similarity query is selected by the electronic device based on a similarity analysis result with the user query vector by a second similarity analysis model;
and the first similarity analysis model and the second similarity analysis model are different analysis models.
The method of claim 1,
The step of selecting a similar query from among the similar query candidates based on the similarity analysis result with the user query vector includes:
determining the degree of similarity between the vector of the similarity query candidate and the user query vector using a similarity analysis model;
The similarity analysis model is a search method learned using a Siamese neural network.
5. The method of claim 4,
The similarity analysis model is trained using the first training data and the second training data,
The first training data consists of a first query pair and a label indicating that the first query pair is similar,
The second training data includes a second query pair and a label indicating that the second query pair is dissimilar.
The method of claim 1,
Receiving a similar query candidate selected based on the similarity with the user query vector from the server,
and receiving a query response corresponding to the similar query candidate.
The method of claim 1,
Receiving a similar query candidate selected based on the similarity with the user query vector from the server,
and receiving a degree of similarity between the vector of the similar query candidate and the user query vector.
The method of claim 1,
The step of selecting a similar query from among the similar query candidates based on the similarity analysis result with the user query vector includes:
determining a similarity ranking of a candidate query belonging to the similar query candidate based on a similarity analysis result with the user query vector; and
and selecting a predetermined number of candidate queries as similar queries according to the similarity ranking.
The method of claim 1,
Receiving a similar query candidate selected based on the similarity with the user query vector from the server,
and receiving a query response corresponding to the similar query candidate.
The method of claim 1,
transmitting the selected similar query to the server; and
The search method further comprising the step of receiving a response to the similar query from the server.
processor; and
a memory for storing instructions executed by the processor;
By executing the instruction, the processor
create a user query vector for the user query,
sending the user query vector to the server;
receiving a similar query candidate selected based on the similarity with the user query vector from the server;
selecting a similar query from among the similar query candidates based on the similarity analysis result with the user query vector;
a search device for determining a response to the similar query as a response to the user query.
generating a user query vector for the user query;
transmitting the user query vector to a server;
receiving, from the server, a similar query candidate selected based on the similarity with the user query vector;
selecting a similar query from among the similar query candidates based on a similarity analysis result with the user query vector; and
A computer-readable recording medium recording a computer program that is executed by a computer, including determining a response to the similar query as a response to the user's query.

Images