KR20210146833A - Apparatus and method for providing summary of document based on genetic algorithm - Google Patents

Abstract

The present disclosure relates to an apparatus and a method for providing the summary of a document based on a genetic algorithm, capable of appropriately selecting an abstract solution while improving a processing speed of the genetic algorithm. According to an embodiment in the present disclosure, an apparatus and a method for providing the summary of a document based on a genetic algorithm includes: a communication circuit to receive an original document from the outside; a memory to store the original document; and a processor electrically connected between the communication circuit and the memory. The processor generates a solution including a plurality of genes corresponding to sentences, which are included in the original document, and are to be analyzed by using the genetic algorithm, and representing some sentences selected to generate the abstract from the sentences to be analyzed, such that the solution satisfies a condition for a preset length of the abstract, calculates a coverage for the solution by using a first object function associated with a similarity between the sentences, which are included in the original document, to be analyzed and the selected some sentences corresponding to the solution, calculates the diversity for the solution by using a second objective function associated with the similarity between the some sentences selected to correspond to the solution, and provides at least one abstract corresponding to each of at least solution satisfying a specific optimal condition based on the coverage and the diversity for each of a plurality of solutions generated through the genetic algorithm.

Description

APPARATUS AND METHOD FOR PROVIDING SUMMARY OF DOCUMENT BASED ON GENETIC ALGORITHM

Embodiments disclosed in this document relate to an apparatus and method for providing a summary for a document using a genetic algorithm.

Due to the development of the ICT field, the amount of information shared with humans has increased significantly and rapidly. With the development of technology, information sharing activities among people have become easier, but ironically, individuals spend more time filtering out false or false information and discovering appropriate content from high-quality information sources. Various automation technologies have been researched and developed to reduce time wasted by individuals, but they do not sufficiently relieve the burden on users.

A system that provides a summary of textual information may be required to quickly filter out unnecessary information and discover suitable content. By using the summary by an automated method, the user can more effectively screen the overflowing information.

In order to provide a summary of the document, various algorithms for extracting the summary may be utilized, and various techniques for evaluating the extracted summary may be utilized. When a genetic algorithm is used to extract abstracts, a very long processing time may be required because there are inevitably too many solutions to be considered. On the other hand, it is necessary to set factors that can clearly reflect the quality of the summary text when evaluating the summary text, and it is necessary to consider various factors in a balanced way.

SUMMARY Embodiments of the present invention provide an apparatus and method for improving the processing speed of a genetic algorithm for generating a summary text and enabling appropriate selection of a summary text solution.

An apparatus for providing a summary of a document based on a genetic algorithm according to an embodiment of the present disclosure includes a communication circuit for receiving an original document from the outside, a memory for storing the original document, and a processor electrically connected to the communication circuit and the memory and the processor satisfies a preset summary sentence length condition by using a genetic algorithm to provide a solution that includes a plurality of genes corresponding to each of the analysis target sentences included in the original document and represents some sentences selected for generation of a summary sentence from the analysis target sentence and calculating the coverage for the solution by using the first objective function associated with the similarity between the analysis target sentences included in the original document and the selected partial sentences corresponding to the solution, and the similarity between the selected partial sentences corresponding to the solution Calculate diversity for a solution by using a second objective function associated with One or more summaries may be provided.

According to an embodiment, when the processor initializes the solution using the genetic algorithm, the processor sequentially sets the values of a plurality of genes included in the solution, and the values of the plurality of genes are based on a preset summary length condition. It is possible to adjust the probability value used to determine .

According to an embodiment, when performing the mutation of the solution using the genetic algorithm, the processor determines whether a plurality of genes included in the solution are mutated based on a preset summary length condition and information on the length of the summary corresponding to the solution. can

According to an embodiment, when the processor performs crossing of a solution and another solution using a genetic algorithm, based on length information of a summary sentence corresponding to a gene selected from a solution and length information of a summary sentence corresponding to a gene selected from another solution Thus, it is possible to decide whether to perform the intersection.

According to an embodiment, the coverage is a degree of similarity between a vector corresponding to an analysis target sentence and a vector corresponding to a selected partial sentence, and a degree of similarity between a vector corresponding to an analysis target sentence and each of a plurality of sentence vectors corresponding to each of the selected partial sentences. can be calculated based on

According to an embodiment, the coverage may be calculated based on the importance of words included in some selected sentences calculated based on a pre-stored knowledge-based database.

According to an embodiment, the coverage may be calculated based on the number of words included in the topic keyword corresponding to some selected sentences and the distribution density of words included in the topic keyword.

According to an embodiment, diversity may increase as the similarity between selected partial sentences decreases.

According to an embodiment, the processor may obtain one or more solutions in which coverage and diversity satisfy a specified condition according to the Pareto optimum, and generate one or more summaries corresponding to each of the one or more solutions.

According to an embodiment, when a solution that does not satisfy the preset summary sentence length condition is generated, the processor determines whether the generated solution satisfies the specified optimization condition, the length of the summary text corresponding to the generated solution is exceeded. Penalties may apply.

The method for providing a summary sentence of a document based on a genetic algorithm according to an embodiment disclosed in this document includes a plurality of genes corresponding to each analysis target sentence included in an original document using a genetic algorithm, and generates a summary sentence from the analysis target sentence generating a solution representing the selected partial sentences to satisfy a preset summary sentence length condition; using a first objective function associated with the similarity between the analysis target sentences included in the original document and the selected partial sentences corresponding to the solution calculating the coverage for the solution, calculating the diversity for the solution by using a second objective function associated with the similarity between some selected sentences corresponding to the solution, and for each of the plurality of solutions generated by the genetic algorithm and providing one or more summaries corresponding to each of one or more solutions satisfying a specified optimization condition based on coverage and diversity.

In the computer recording medium storing instructions executable by at least one processor included in the computing device according to an embodiment disclosed in this document, the instructions cause the at least one processor to add to the original document using a genetic algorithm. A solution including a plurality of genes corresponding to each included analysis target sentence and representing some sentences selected for generation of a summary sentence from the analysis target sentence is generated to satisfy a preset summary sentence length condition, and the analysis target sentence included in the original document A coverage for the solution is calculated using a first objective function associated with the degree of similarity between the selected partial sentences corresponding to the solution, and a second objective function associated with the similarity between the selected partial sentences corresponding to the solution is used for the solution. The diversity may be calculated, and one or more summaries corresponding to each of the one or more solutions satisfying a specified optimization condition may be provided based on the coverage and diversity of each of the plurality of solutions generated by the genetic algorithm.

According to the embodiments disclosed in this document, the processing time for generating the summary can be shortened by performing the genetic algorithm process (eg, initialization, mutation, crossover, etc.) in consideration of the preset length condition of the summary.

In addition, by performing multi-objective optimization using two objective functions, it is possible to efficiently evaluate solutions and select excellent summaries.

In addition, by selecting one or more summary sentences that satisfy a certain quality or more, it is possible to select an appropriate summary according to the user's intention.

In addition, various effects directly or indirectly identified through this document may be provided.

1 illustrates an exemplary summary provided by an apparatus for providing a summary of a document based on a genetic algorithm according to an embodiment.
2 is a block diagram illustrating a configuration of an apparatus for providing a document summary based on a genetic algorithm according to an exemplary embodiment.
3 is a diagram for explaining an exemplary operation of an apparatus for providing a summary of a document based on a genetic algorithm according to an embodiment.
4 is a diagram for explaining an exemplary operation of an apparatus for providing a summary text of a document based on a genetic algorithm according to an embodiment.
5 is a diagram for explaining an exemplary operation of an apparatus for providing a document summary based on a genetic algorithm according to an embodiment.
6 is a diagram for explaining an exemplary operation of an apparatus for providing a summary text of a document based on a genetic algorithm according to an embodiment.
7 is a diagram for describing an exemplary operation of an apparatus for providing a summary text of a document based on a genetic algorithm according to an embodiment.
8 is a diagram for explaining an exemplary operation of an apparatus for providing a summary text of a document based on a genetic algorithm according to an embodiment.
9 is a flowchart illustrating a method of providing a document summary based on a genetic algorithm according to an embodiment.
In connection with the description of the drawings, the same or similar reference numerals may be used for the same or similar components.

Hereinafter, some embodiments of the present invention will be described in detail with reference to exemplary drawings. However, this is not intended to limit the present invention to specific embodiments, and it should be understood that various modifications, equivalents or substitutes of the embodiments of the present invention are included. In adding reference numerals to the components of each drawing, it should be noted that the same components are given the same reference numerals as much as possible even though they are indicated on different drawings. In addition, in describing the embodiment of the present invention, if it is determined that a detailed description of a related known configuration or function interferes with the understanding of the embodiment of the present invention, the detailed description thereof will be omitted.

1 illustrates an exemplary summary provided by an apparatus for providing a summary of a document based on a genetic algorithm according to an embodiment.

The summary text providing apparatus according to an embodiment may generate the summary text 120 by extracting some of the sentences included in the plurality of original documents 111 , 112 , 113 , and 114 .

For example, the summary providing apparatus may acquire the plurality of original documents 111 , 112 , 113 , and 114 . The apparatus for providing a summary may derive a solution corresponding to some sentences to be included in the summary 120 among sentences included in the plurality of original documents 111 , 112 , 113 and 114 by using a genetic algorithm. The summary text providing apparatus inputs the derived solution into two objective functions, thereby providing coverage related to the similarity between the plurality of original documents 111 , 112 , 113 , and 114 and the summary text 120 and sentences included in the summary text 120 . Diversity associated with the degree of similarity may be calculated. It can be determined whether the summary text 120 sufficiently reflects the contents of the original documents 111, 112, 113, and 114 by the coverage, and whether unnecessary sentences are included in the summary text 120 can be determined by the diversity. can The apparatus for providing a summary may derive a solution in which coverage and diversity satisfy optimization conditions, and may provide a summary 120 corresponding to the derived solution.

The examples described above are illustratively described for the understanding of this document, and the scope of the rights of this document is not limited thereto. Hereinafter, a specific method for using a genetic algorithm, evaluating a solution by multi-purpose optimization, and providing a summary will be described in detail.

2 is a block diagram illustrating a configuration of an apparatus for providing a document summary based on a genetic algorithm according to an exemplary embodiment.

Referring to FIG. 2 , an apparatus 200 for providing a document summary based on a genetic algorithm according to an embodiment may include a communication circuit 210 , a memory 220 , and a processor 230 . The summary providing apparatus 200 may be, for example, a computing device such as a desktop, a laptop, a tablet, or a smart phone as a user terminal, or may be implemented in the form of a server. In this case, it may be physically located on premise or in the cloud. As another example, the summary providing apparatus 200 may be implemented as a computing device (eg, one user terminal and one server) in a distributed environment in two or more. The summary providing apparatus 200 is for efficiently providing a summary of a document, but is not limited thereto, and may be utilized to provide a summary of a single document.

The communication circuit 210 may be configured to communicate with the outside wirelessly or by wire. The communication circuit 210 may transmit/receive data to and from an external device. For example, the communication circuit 210 may receive one or more original documents from the outside.

Memory 220 may include volatile memory and/or non-volatile memory. The memory 220 may store various data handled by the apparatus 200 for providing a summary of a document. For example, the memory 220 may store data processed inside the apparatus 200 for providing a summary of a document, or may store data received from the outside. For example, the memory 220 may store one or more received original documents.

The processor 230 may be electrically connected to the communication circuit 210 and the memory 220 . The processor 230 may control the communication circuit 210 and the memory 220 , and may perform various data processing and operations. Although the processor 230 is illustrated as a single configuration in FIG. 2 , it may be implemented as a plurality of separate components. The processor 230 may provide the user with a tool for providing a summary of the document. The processor 230 may perform the following operations by executing software or instructions stored in the memory 220 .

According to an embodiment, the processor 230 uses a genetic algorithm to include a plurality of genes corresponding to each of the analysis target sentences included in the original document and provides a solution representing some sentences selected for generating a summary sentence from the analysis target sentence. It can be created to satisfy the preset summary length condition. In this document, the term “sentence” may be interpreted to mean one sentence, a part of a sentence (such as a phrase or word including two or more words), or a set of two or more sentences. The analysis target sentence may mean the entire sentence included in the original document, or may mean a sentence excluding some sentences from the whole. A solution of the genetic algorithm may be a vector containing a gene (component) consisting of binary data. The solution vector may indicate which sentences to be included in the candidate summary sentence among the analysis target sentences included in the original document. For example, if the original document consists of 10 sentences, and the candidate summary contains the 1st, 3rd, and 7th sentences, the solution vector is (1, 0, 1, 0, 0, 0, 1, 0) , 0, 0). The processor 230 may generate a plurality of (or a plurality of) solutions corresponding to a plurality of (or a plurality of) candidate summaries by performing processes such as initialization, mutation, and crossover using a genetic algorithm. The summary sentence length condition is the number of words (eg 100 words or less), the number of sentences (eg 10 sentences or less), or the length of the summary compared to the length of the original document (eg 10% of the number of sentences in the original document or the number of words in the original document) 10%) and the like. In this document, the term “word” may consist of text or data composed of numbers, vectors, matrices, or tensors. Also, the term “word” may be understood as a concept including a subword. The processor 230 may generate a solution corresponding to the candidate summary text to satisfy a preset summary text length condition when performing processes such as initialization, mutation, and intersection of the solution. The initialization, mutation, and crossover of a solution using a genetic algorithm will be described in detail with reference to FIGS. 3 to 7 .

According to an embodiment, the processor 230 sequentially sets the values of a plurality of genes included in the solution when the solution is initialized using a genetic algorithm, but based on a preset summary length condition, the plurality of The probability value used to determine the value of a gene can be adjusted. When initialization is performed using a genetic algorithm for a solution made of binary data, a new solution may be generated by sequentially setting each gene included in the solution to 0 or 1 by random probability. In the case of deriving a solution for the summary text, if the length of the summary text corresponding to the solution becomes excessively long, the summary text may lose its original function. Candidate summaries having an excessively long length are not only less necessary, but may require unnecessary computation in the system. Therefore, according to the condition limiting the length of the candidate summary, it is possible to limit the solution generated by initialization. When the solution is initialized, the processor 230 sequentially generates 0 or 1 genes to be included in the solution with a random probability, but the length of the candidate summary is used to determine the value of the gene as 0 or 1 so that the length condition is satisfied. It is possible to adjust the probability value of the genetic algorithm. As another example, the processor 230 may set values of the remaining genes based on a preset summary length condition and length information of a summary text corresponding to the preset gene. When the solution is initialized, the processor 230 sequentially generates 0 or 1 genes to be included in the solution with random probability, but the length of the candidate summary corresponding to the already generated gene reaches the length condition (or exceeds the length condition) ), the remaining genes that were not generated may be generated as 0, or the probability of being generated to 0 may be increased. The initialization of the solution using the genetic algorithm will be described in detail with reference to FIGS. 3 to 5 .

According to an embodiment, when the processor 230 performs the mutation of the solution using the genetic algorithm, mutation of a plurality of genes included in the solution based on a preset summary length condition and length information of the summary corresponding to the solution can decide whether When mutation is performed using a genetic algorithm on a solution made of binary data, another solution may be generated by changing at least one of the genes included in the solution with random probability. For example, gene 0 may be mutated to 1, and gene 1 may be mutated to 0. In this case, especially when gene 0 is mutated to 1, the length of the candidate summary corresponding to the solution generated by the mutation may exceed the length condition. Therefore, according to the condition limiting the length of the candidate summary, the performance of the mutation may be restricted. The processor 230 mutates at least one of the genes included in the solution to 0 or 1 with a random probability, and when the length of the candidate summary corresponding to the solution generated by the mutation exceeds the length condition, The probability of performing mutation may be reduced or mutation may not be performed. The variation of the solution using the genetic algorithm will be described in detail with reference to FIG. 6 .

According to an embodiment, when the processor 230 crosses a solution and another solution using a genetic algorithm, information on the length of a summary text corresponding to the gene selected from the solution and the length of the summary text corresponding to the gene selected from another solution Based on the information, it is possible to determine whether to perform the intersection. When crossing two solutions made of binary data using a genetic algorithm, a part of the first solution and a part of the second solution may be combined with an arbitrary probability. For example, intersecting a first solution having ten components and a second solution results in the first through fifth components of the first solution and the sixth through tenth components of the second solution being combined, resulting in another solution can be created In this case, the length of the candidate summary corresponding to the solution generated by the intersection may exceed the length condition. Therefore, according to the condition limiting the length of the candidate summary, the performance of the intersection can be restricted. The processor 230 selects some of the genes included in the two solutions and crosses them by random probability, and when the length of the candidate summary corresponding to the solution generated by the crossing exceeds the length condition, the two solutions It is possible to reduce the probability of performing the intersection for , or not to perform the intersection. The intersection of a solution using a genetic algorithm will be described in detail with reference to FIG. 7 .

According to an embodiment, when generating a solution corresponding to a candidate summary by using a genetic algorithm, the processor 230 may limit the number of generations. Also, the processor 230 may stop generating the next generation when there is no performance improvement of the candidate summary. In addition, the processor 230 may store the current progress, retrieve the stored result later, and then resume generation of the next generation.

According to an embodiment, the processor 230 may calculate the coverage for the solution by using the first objective function associated with the similarity between the analysis target sentences included in the original document and some selected sentences corresponding to the solution. By utilizing the similarity between the sentences included in the original document and the candidate summary corresponding to a specific solution, coverage indicating how well the candidate summary reflects the content of the original document may be calculated. For example, the coverage is based on the degree of similarity between the vector corresponding to the analysis target sentence and the vector corresponding to the selected partial sentence and the similarity between the vector corresponding to the analysis target sentence and each of the plurality of sentence vectors corresponding to each of the selected partial sentences. can be calculated by A vector (a vector corresponding to the original document) corresponding to the analysis target sentence of the original document may be expressed by the following exemplary equation.

[Equation 1]

Here, o is a vector corresponding to the original document, o _k is a value (or vector) corresponding to the k-th sentence of the original document, w _ik is a weight corresponding to the i-th word of the k-th sentence, and w _ik is It can be calculated by TF-ISF (term frequency-inverse sentence frequency).

Meanwhile, a vector corresponding to a sentence included in the candidate summary sentence corresponding to the solution (a vector corresponding to the candidate summary sentence) and a sentence vector for each sentence included in the candidate summary sentence may be expressed by the following exemplary equations.

[Equation 2]

Here, s _i is the sentence vector of the i-th sentence included in the original document, w _im is the weight corresponding to the m-th word included in the i-th sentence, o ^s is a vector corresponding to the candidate summary, o ^s _k is a value (or vector) corresponding to the m-th sentence included in the candidate summary sentence, and x _i may be the i-th gene (0 or 1) of the solution vector.

An exemplary equation (first objective function) for calculating coverage by using the vector corresponding to the above-described original document, the vector corresponding to the candidate summary, and the sentence vector for each sentence included in the candidate summary is as follows.

[Equation 3]

Here, n may be the number of analysis target sentences included in the original document, and sim may be a function for calculating the similarity and may be a cosine similarity function. A coverage reflecting the similarity between the original document and the candidate summary corresponding to the specific solution and the similarity between each of the sentences included in the original document and the candidate summary may be calculated by the above-described formula. The first objective function for calculating the coverage may be normalized by dividing by the average length value of sentences (eg, sentences included in the candidate summary) involved in calculating the coverage.

According to an embodiment, the coverage is calculated based on the importance of words included in some selected sentences calculated based on a pre-stored knowledge-based database (including various types of data such as a file in a dictionary format) it might be The importance of a word calculated based on the knowledge-based database and the coverage taking it into consideration may be calculated by the following exemplary equation. By considering the importance of words included in the candidate summary, the quality of the candidate summary may be more accurately determined.

[Equation 4]

Here, μ ^s _k is the importance of the k-th sentence included in the candidate summary, and y _i is the value (eg, binary value) of the i-th word of the k-th sentence obtained from the knowledge base (eg, DBpedia) have.

According to an embodiment, the coverage may be calculated based on the number of words included in the topic keyword corresponding to some selected sentences and the distribution density of words included in the topic keyword. The coverage in consideration of the topic keyword may be calculated by the following exemplary equation. By considering the topic keywords of sentences included in the candidate summary, the quality of the candidate summary may be more accurately determined.

[Equation 5]

Here, Ii is the amount of increase in the number of words included in the topic keyword of the i-th sentence compared to the i-1th sentence, Di is the amount of decrease in the number of words included in the topic keyword of the i-th sentence compared to the i-1th sentence, and Si is It may be a distribution density of words used in calculating the topic keyword.

According to an embodiment, the processor 230 may calculate diversity for the solution by using the second objective function associated with the similarity between some selected sentences corresponding to the solution. Diversity may be calculated to increase as the similarity between some selected sentences decreases. By utilizing the similarity between sentences included in the candidate summary corresponding to a specific solution, diversity for determining how many unnecessary sentences are included in the candidate summary may be calculated. For example, the diversity may be calculated based on the similarity between sentence vectors included in the candidate summary. An exemplary equation (second objective function) for calculating diversity using a sentence vector included in a candidate summary sentence is as follows.

[Equation 6]

Here, s _i may be the i-th sentence included in the original document, and s _j may be the j-th sentence included in the original document. _{By using x i} and x _j , which are components of the solution vector, the similarity of sentences included in the candidate summary can be considered. Diversity reflecting the similarity between one sentence and another sentence included in the candidate summary sentence may be calculated by the above-described equation. The second objective function for calculating diversity may be normalized by dividing by the average length value of sentences (eg, sentences included in candidate summary sentences) involved in calculating coverage.

According to an embodiment, the processor 230 may provide one or more summaries corresponding to each of one or more solutions that satisfy a specified optimization condition based on the coverage and diversity of each of the plurality of solutions generated by the genetic algorithm. have. For example, the processor 230 may obtain one or more solutions in which coverage and diversity satisfy a specified condition according to a Pareto optimum. The processor 230 may generate one or more summaries corresponding to each of the one or more solutions, and provide the generated one or more summaries to the user. When a plurality of solutions satisfying the specified optimization condition are obtained, the processor 230 may derive one solution through additional processing and provide one summary text corresponding thereto to the user, or a plurality of solutions corresponding to each of the plurality of solutions. It is also possible to give the user a choice by providing the user with a summary of the

According to an embodiment, when a solution that does not satisfy the preset summary length condition is generated, the processor 230 determines whether the generated solution satisfies the specified optimization condition, the summary text corresponding to the generated solution exceeds the A penalty according to one length may be applied. When the summary sentence length condition is set to be short, the coverage and diversity values of the solution exceeding the summary length condition may be superior to the coverage and diversity values of the solution satisfying the abstract length condition. Therefore, in some cases, it is necessary to consider a solution that does not satisfy the summary sentence length condition. The penalty may be calculated, for example, as a ratio (overflow/limit) between the length exceeded and the specified summary length.

3 is a diagram for explaining an exemplary operation of an apparatus for providing a summary of a document based on a genetic algorithm according to an embodiment.

Referring to FIG. 3 , when the apparatus for providing a summary text according to an embodiment performs initialization of a solution using a genetic algorithm, values of a plurality of genes included in the solution are sequentially set, but based on a preset summary text length condition Thus, the probability value used to determine the values of the plurality of genes may be adjusted.

For example, when there are 10 analysis target sentences in the original document, the summary sentence providing apparatus may generate a chromosome (solution) including 10 genes. When a chromosome is initialized, it may be sequentially generated to have a value of 0 or 1 from the 1st gene to the 10th gene. The summary sentence length condition may be set to, for example, three sentences. The summary providing apparatus may adjust the probability value used for gene setting so that the probability that the length of the candidate summary is set to 3 increases. The apparatus for providing a summary sentence may generate a chromosome such that the length of the candidate summary text approaches 3 by adjusting σ according to the summary text length condition.

As another example, when the 1st gene is set to 1, the 2nd gene is set to 0, the 3rd gene is set to 0, the 4th gene is set to 1, and the 5th gene is set to 1, the candidate summary corresponding to the preset gene can contain 3 sentences. As the number of sentences included in the candidate summary corresponding to the preset gene increases, the probability that the gene is set to 1 may decrease. In this case, if even one of the 6th to 10th genes is set to 1, the generated chromosome may not satisfy the summary length condition. Accordingly, the 6th gene to the 10th gene may be set to 0. This allows chromosomes to be initialized to correspond to candidate summaries of length 3 or less.

4 is a diagram for explaining an exemplary operation of an apparatus for providing a summary text of a document based on a genetic algorithm according to an embodiment.

Referring to FIG. 4 , when the apparatus for providing a summary text according to an embodiment performs initialization of a solution using a genetic algorithm, the values of a plurality of genes included in the solution are sequentially set, but based on a preset summary text length condition Thus, the probability value used to determine the values of the plurality of genes may be adjusted.

For example, when there are 10 analysis target sentences in the original document, the summary sentence providing apparatus may generate a chromosome (solution) including 10 genes. When a chromosome is initialized, it may be sequentially generated to have a value of 0 or 1 from the 1st gene to the 10th gene. The summary sentence length condition may be set as a range, for example, 3 to 4 sentences. In this case, chromosomes may be generated such that the candidate summary preferably contains a minimum of 3 sentences and a maximum of 4 sentences. The summary providing apparatus may adjust the probability value used for gene setting so that the probability that the length of the candidate summary is set to 3 to 4 increases. The apparatus for providing a summary may generate chromosomes such that the length of the candidate summary approaches 3 to 4 by arbitrarily adjusting σ to α (eg, 3) to β (eg, 4) according to the summary length condition.

As another example, when the length of the candidate summary corresponding to the preset gene is shorter than the minimum length, the gene may be set to 1 with a relatively high probability (eg, when the first gene to the fifth gene is generated), and the candidate If the length of the summary is longer than the minimum length and shorter than the maximum length, there is a relatively low probability that a gene can be set to 1 (eg, when generating 6th to 9th genes). When the length of the candidate summary corresponding to the preset gene reaches the maximum length, the remaining genes may be set to 0 (eg, when the 10th gene is created). This allows chromosomes to be initialized to correspond to candidate summaries of length 3 or greater and 4 or less.

5 is a diagram for explaining an exemplary operation of an apparatus for providing a document summary based on a genetic algorithm according to an embodiment.

FIG. 5A is a graph illustrating a relationship between coverage and diversity of a generated solution when initialization is performed without applying a summary sentence length condition. FIG. 5B is a graph illustrating a relationship between coverage and diversity of a generated solution when initialization is performed by applying a summary sentence length condition.

When the length condition of the summary is not applied, a plurality of candidate summary sentences having low diversity may be generated because a plurality of unnecessary sentences may be included. Meanwhile, when the length condition of the summary sentence is applied, candidate summary sentences having various distributions of coverage and diversity may be generated.

6 is a diagram for explaining an exemplary operation of an apparatus for providing a summary text of a document based on a genetic algorithm according to an embodiment.

Referring to FIG. 6 , when the apparatus for providing a summary text according to an embodiment performs a solution variation using a genetic algorithm, a plurality of pieces included in a solution based on a preset summary text length condition and length information of a summary text corresponding to the solution can determine whether the gene is mutated.

For example, the summary providing apparatus may perform mutation on a chromosome (solution) including 10 genes. When a gene included in a chromosome is mutated, the gene value may be mutated according to an arbitrary probability. The summary providing apparatus may adjust the probability value used for genetic mutation based on the ratio of 0, the ratio of 1, and the summary length condition included in the chromosome. The summary sentence length condition may be set to, for example, three sentences. When a chromosome includes two genes with a value of 1, the probability that a gene with a value of 0 will be mutated into 1 may be set to be relatively high. When a chromosome includes three genes with a value of 1, the probability that a gene with a value of 0 will be mutated into 1 may be set to be relatively low. Thereby, the chromosome can be mutated so that the length of the candidate summary is close to three.

For another example, if a chromosome contains two genes with a value of 1, the 7th gene with a value of 0 may be mutated to 1. If a chromosome contains three genes with a value of 1, The 7th gene may not be mutated. This allows chromosomes to be mutated to correspond to candidate summaries of length 3 or less.

7 is a diagram for describing an exemplary operation of an apparatus for providing a summary text of a document based on a genetic algorithm according to an embodiment.

Referring to FIG. 7 , the apparatus for providing a summary according to an embodiment corresponds to length information of a summary text corresponding to a gene selected from a solution and a gene selected from another solution when crossing a solution with another solution using a genetic algorithm Based on the length information of the summary sentence, it is possible to determine whether to perform the intersection.

For example, the summary providing device may perform crossover on chromosomes (solutions) comprising 10 genes. In the case of crossing chromosomes, the crossing point and whether the chromosomes are crossed may be determined according to an arbitrary probability. The summary providing apparatus may adjust the probability value used for crossover based on the ratio of 0, the ratio of 1, and the summary length condition included in the chromosome. The summary sentence length condition may be set to, for example, three sentences. When the chromosome to be crossed includes three genes having a value of 1, the probability of performing the crossover may be set to be relatively high. When the chromosome to be crossed includes four genes having a value of 1, the probability of performing the crossover may be set to be relatively low. This allows chromosomes to be crossed so that the length of the candidate summary is close to three.

As another example, crossover may be performed when the chromosome to be crossed contains 3 genes with a value of 1, and crossover may not be performed when the chromosome to be crossed contains 4 genes with a value of 1. have. This allows chromosomes to be crossed to correspond to candidate summaries of length 3 or less.

8 is a diagram for explaining an exemplary operation of an apparatus for providing a summary text of a document based on a genetic algorithm according to an embodiment.

Referring to FIG. 8 , the apparatus for providing a summary according to an embodiment may acquire one or more solutions satisfying a specified optimization condition based on coverage and diversity of each of a plurality of solutions. A summary sentence can perform its original function when it properly reflects the content of the original document and does not contain unnecessary sentences. Accordingly, by performing multi-object optimization on the first objective function and the second objective function, it is possible to derive a high-quality summary.

The graph shown in FIG. 8 may represent a relationship between the coverage and diversity of a solution that satisfies the Pareto optimal condition among solutions generated by the genetic algorithm. The apparatus for providing a summary may select one or more solutions in which coverage and diversity satisfy optimization conditions. The apparatus for providing a summary may select a solution having high coverage, a solution having high diversity, or a solution having a balance between coverage and diversity among solutions that satisfy the optimization condition. The summary text providing apparatus may provide a plurality of solutions to the user and induce the user's selection.

9 is a flowchart illustrating a method of providing a document summary based on a genetic algorithm according to an embodiment.

Hereinafter, it is assumed that the apparatus 200 for providing a summary of a document of FIG. 2 performs the process of FIG. 9 . Also, in the description of FIG. 9 , it may be understood that an operation described as being performed by the apparatus for providing a summary of a document is controlled by the processor 230 .

Referring to FIG. 9 , in step 910, the device includes a plurality of genes corresponding to each of the analysis target sentences included in the original document using a genetic algorithm, and a solution representing some sentences selected for generation of a summary sentence from the analysis target sentence can be generated to satisfy a preset summary sentence length condition.

In operation 920, the device may calculate the coverage for the solution by using the first objective function associated with the similarity between the analysis target sentences included in the original document and some selected sentences corresponding to the solution.

In operation 930, the device may calculate diversity for the solution by using the second objective function associated with the similarity between some selected sentences corresponding to the solution.

In operation 940 , the device may provide one or more summaries corresponding to each of the one or more solutions satisfying the specified optimization condition based on the coverage and diversity for each of the plurality of solutions generated by the genetic algorithm.

The embodiments of this document and the terms used therein are not intended to limit the technology described in this document to a specific embodiment, but it should be understood to include various modifications, equivalents, and/or substitutions of the embodiments. In connection with the description of the drawings, like reference numerals may be used for like components. The singular expression may include the plural expression unless the context clearly dictates otherwise. In this document, expressions such as “A or B”, “at least one of A and/or B”, “A, B or C” or “at least one of A, B and/or C” refer to all of the items listed together. Possible combinations may be included. Expressions such as "first," "second," "first," or "second," can modify the corresponding elements regardless of order or importance, and to distinguish one element from another element. It is used only and does not limit the corresponding components. When a component is referred to as being “connected (functionally or communicatively)” or “connected” to another component, the component is directly connected to the other component or refers to another component. can be connected through

In this document, "adapted to or configured to", depending on the context, for example, hardware or software "suitable for," "having the ability to," "modified to, Can be used interchangeably with ""made to," "capable of," or "designed to." In some circumstances, the expression “a device configured to” may mean that the device is “capable of” with other devices or parts. For example, the phrase "a processor configured (or configured to perform) A, B, and C" refers to a dedicated processor (eg, an embedded processor) for performing the corresponding operations, or by executing one or more programs stored in a memory device; It may refer to a general-purpose processor (eg, CPU) capable of performing corresponding operations.

As used herein, the term “module” includes a unit composed of hardware, software, or firmware, and may be used interchangeably with terms such as, for example, logic, logic block, component, or circuit. can A “module” may be an integrally formed component or a minimum unit or a part that performs one or more functions. A “module” may be implemented mechanically or electronically, for example, known or to be developed, application-specific integrated circuit (ASIC) chips, field-programmable gate arrays (FPGAs), or It may include a programmable logic device.

At least a portion of an apparatus (eg, modules or functions thereof) or a method (eg, operations) according to an embodiment may be implemented as instructions stored in a computer-readable storage medium in the form of a program module. When the instruction is executed by the processor, the processor may perform a function corresponding to the instruction.

Each of the components (eg, a module or a program module) according to an embodiment may be composed of a singular or a plurality of entities, and some sub-components of the aforementioned sub-components may be omitted or other sub-components may be included. may include more. Alternatively or additionally, some components (eg, a module or a program module) may be integrated into one entity to perform the same or similar functions performed by each corresponding component before being integrated. Operations performed by a module, program module, or other component according to an embodiment are sequentially, parallelly, repetitively or heuristically executed, or at least some operations are executed in a different order, omitted, or other operations This can be added.

Claims (12)

In the apparatus for providing a summary of a document based on a genetic algorithm,
a communication circuit for receiving an original document from the outside;
a memory for storing the original document; and
a processor electrically coupled to the communication circuitry and the memory;
The processor is
Using a genetic algorithm, a solution including a plurality of genes corresponding to each analysis target sentence included in the original document and representing some sentences selected for generating a summary sentence from the analysis target sentence is generated to satisfy a preset summary sentence length condition do,
Calculating the coverage for the solution by using a first objective function associated with the similarity between the analysis target sentence included in the original document and the selected partial sentence corresponding to the solution,
calculating diversity for the solution by using a second objective function associated with the similarity between the selected partial sentences corresponding to the solution;
and providing one or more summaries corresponding to each of one or more solutions satisfying a specified optimization condition based on the diversity and the coverage for each of a plurality of solutions generated by the genetic algorithm. The method of claim 1,
The processor is
When initialization of the solution is performed using the genetic algorithm, the values of the plurality of genes included in the solution are sequentially set, and the values of the plurality of genes are determined based on the preset summary sentence length condition A device, characterized in that for adjusting the probability value used for The method of claim 1,
The processor is
When mutation of the solution is performed using the genetic algorithm, determining whether the plurality of genes included in the solution are mutated based on the preset summary length condition and length information of the summary text corresponding to the solution Characterized by the device. The method of claim 1,
The processor is
When the intersection of the solution and another solution is performed using the genetic algorithm, the crossover is performed based on length information of the summary corresponding to the gene selected in the solution and the length information of the summary corresponding to the gene selected in the other solution. A device, characterized in that determining whether to perform. The method of claim 1,
The coverage is based on the degree of similarity between the vector corresponding to the analysis target sentence and the vector corresponding to the selected partial sentence, and the similarity between the vector corresponding to the analysis target sentence and each of the plurality of sentence vectors corresponding to each of the selected partial sentences. Device, characterized in that calculated on the basis of. 6. The method of claim 5,
The apparatus, characterized in that the coverage is calculated based on the importance of words included in the selected partial sentences calculated based on a pre-stored knowledge-based database. 6. The method of claim 5,
The apparatus, characterized in that the coverage is calculated based on the number of words included in the topic keyword corresponding to the selected partial sentences and the distribution density of words included in the topic keyword. The method of claim 1,
The apparatus, characterized in that the diversity increases as the degree of similarity between the selected partial sentences decreases. The method of claim 1,
The processor is
obtaining the one or more solutions in which the coverage and the diversity satisfy a specified condition according to a Pareto optimum;
and generating one or more summaries corresponding to each of the one or more solutions. The method of claim 1,
The processor is
When a solution that does not satisfy the preset summary length condition is generated, a penalty is applied according to the length exceeding the summary text corresponding to the generated solution when determining whether the generated solution satisfies the specified optimization condition A device, characterized in that. A method for providing a summary of a document based on a genetic algorithm, the method comprising:
Generating a solution that includes a plurality of genes corresponding to each of the analysis target sentences included in the original document using a genetic algorithm and that represents some sentences selected for generation of a summary sentence from the analysis target sentence to satisfy a preset summary sentence length condition step;
calculating coverage for the solution by using a first objective function associated with a degree of similarity between the analysis target sentence included in the original document and the selected partial sentence corresponding to the solution;
calculating diversity for the solution by using a second objective function associated with a similarity between the selected partial sentences corresponding to the solution; and
providing one or more summaries corresponding to each of one or more solutions satisfying a specified optimization condition based on the diversity and the coverage for each of a plurality of solutions generated by the genetic algorithm; Way. In a computer recording medium storing instructions executable by at least one processor included in a computing device,
The instructions cause the at least one processor,
Using a genetic algorithm, a solution including a plurality of genes corresponding to each analysis target sentence included in the original document and representing some sentences selected for generating a summary sentence from the analysis target sentence is generated to satisfy a preset summary sentence length condition, ,
Calculating the coverage for the solution by using a first objective function associated with the similarity between the analysis target sentence included in the original document and the selected partial sentence corresponding to the solution,
calculating diversity for the solution by using a second objective function associated with the similarity between the selected partial sentences corresponding to the solution;
and to provide one or more summaries corresponding to each of one or more solutions satisfying a specified optimization condition based on the diversity and the coverage for each of the plurality of solutions generated by the genetic algorithm. .

Images