KR20130127804A - Apparatus and method for document summary based on query - Google Patents

Abstract

The present invention discloses a query-based document summarization device and a method thereof. Given in the present invention, the query-based document summarization device includes a preprocessor which segments an input document into sentences and generates a matrix of the sentences; a query extender which extends queries using the intent-related feedback; a weighting calculator which re-calculates the weighting of a term using the semantic characteristics through non-negative matrix factorization; and a summarizer which generates summarized sentences for the input document using the matrix of the terms and the sentences into which the extended user queries and the re-calculated term weightings are applied. [Reference numerals] (110) Preprocessor;(120) Question extender;(130) Weighed value calculator;(140) Sentence summary device

Description

Apparatus and Method for Document Summary based on Query}

The present invention relates to a summary sentence generating technique, and more particularly to a query-based document summary apparatus and method that can generate a summary reflecting the user's intention.

Recently, the amount of information continues to explode with the development of the Internet, and such information can be easily accessed from anywhere due to the development of information communication and portable terminals.

However, a large amount of information also causes a user to not easily access the desired information. For this reason, there is an increasing need for a summary that enables users to easily check the desired information, and accordingly, a lot of researches on document summaries have been conducted.

The document summary can be classified as follows according to the purpose of the summary. First, there is a general summary or comprehensive summary that summarizes the entire contents of the document. Second, there is a query-based document summary that summarizes the document with the contents of the user's query. Third, there are multiple document summaries summarizing the contents related to the same topic from several documents, and single document summaries summarizing related topics from a single document. Finally, there is a personalized document summary that summarizes the user log on the Internet and special information related to the user's needs to suit the individual's characteristics. In addition, document summaries can be divided into document summaries using statistical methods, graph-based methods, linguistics-based methods, semantic information-based methods, external resource-based methods, and other complex-based methods, depending on the techniques used in the summary.

The present invention has been made in the technical background as described above, by extracting a sentence according to a user's intention from the input document by means of semantic features and pseudo-related feedback extracted from non-negative matrix decomposition, and to generate a summary sentence using the extracted sentence It is an object of the present invention to provide an apparatus and method for query-based document summarization.

In accordance with an aspect of the present invention, a query-based document summarizing apparatus includes: a preprocessor configured to decompose an input document into sentences to generate a term sentence matrix; A query extender that extends user queries using correlation feedback; A weight calculator for recalculating the weights of terms using nonnegative matrixed semantic features; And a summarizer for generating a summary sentence for the input document by using the extended terminology matrix to which the user query and the weighted term of the redefined term are applied.

According to the present invention, a sentence is extracted by extracting a sentence according to a user's intention from an input document by using a query extended by pseudo-related feedback and a term weight defined by nonnegative matrix decomposition, and thus, a summary sentence is generated. You can create a summary that can be expressed well, and improve the quality of the generated summary.

1 is a block diagram showing a query-based document summary device according to an embodiment of the present invention.

Advantages and features of the present invention and methods for achieving them will be apparent with reference to the embodiments described below in detail with the accompanying drawings. The present invention may, however, be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art. Is provided to fully convey the scope of the invention to those skilled in the art, and the invention is only defined by the scope of the claims. It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. In the present specification, the singular form includes plural forms unless otherwise specified in the specification. As used herein, the terms " comprises, " and / or "comprising" refer to the presence or absence of one or more other components, steps, operations, and / Or additions.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. 1 is a block diagram showing a query-based document summary device according to an embodiment of the present invention.

As shown in FIG. 1, the query based document summary apparatus 10 according to an embodiment of the present invention includes a preprocessor 110, a query expander 120, a weight calculator 130, and a sentence summarizer 140. do.

The preprocessor 110 performs sentence decomposition, stop-word removal, stemming extraction, and term sentence matrix generation for the input document.

In detail, the preprocessor 110 may remove the stopword from the input document by using the stopword list of Rijsbergen and the stopword removal method by lexical analysis.

In addition, the preprocessor 110 may extract the root of the input document using a Porter stepping algorithm.

In this case, the preprocessor 110 may perform the above-described operation when receiving an English document, and may preprocess the input document by using a morphological analysis tool when receiving a Korean document.

The query expander 120 expands the received user's query by using the correlation feedback.

In detail, the related feedback has a large processing load because the user needs to continuously check the result, but the pseudo-related feedback can minimize the interference of the user, and thus, the feedback has been widely used in recent automated information retrieval systems. Accordingly, the query expander 120 of the present invention also applies the correlation feedback to reduce the burden on the user.

Specifically, pseudo-relationship is a method of expanding a query using the top k sentences in order of high similarity with the user query.

In this case, if the number of sentences k used to expand the query is too large, the summary result is too comprehensive and ambiguous for the user's desired subject, and if the number of sentences k is too small, the summary result is too large for the subject requested by the user. It can be narrow. Therefore, it is important to select an appropriate amount of k sentences in the correlational feedback.

However, unlike the related feedback in the related art, since the unrelated document cannot be determined, the number of related feedbacks calculated by Equation 1 below is used.

를 산출한다.The weight calculator 130 substitutes the weight matrix W into the term sentence matrix D as shown in Equation 2 below to determine the weight of the term.

.

는 용어에 대응하는 비음수 행렬분해된 의미특징의 합을 그 값으로 갖는 대각행렬로서, 하기의 수학식 3과 같이 표현될 수 있다.here,

Is a diagonal matrix having, as its value, the sum of non-negative matrix-factored semantic features corresponding to the term, which may be expressed by Equation 3 below.

In Equation 3, w _i is a sum of semantic features for calculating a weight of a term, and may be calculated by a non-negative matrix decomposition algorithm as in Equation 4 below.

Specifically, the sum w _i of the semantic features may be calculated by simultaneously updating the values of the semantic feature matrices W and H until the Euclidean distance J converges close to zero, as shown in equations (5) and (6).

In other words, Equation 5 is for decomposing the matrix A into a non-negative m × r matrix W and a non-negative r × n matrix H. Here, A is an m × n matrix composed of m pixels and n images, and r is the number of semantic features, which is set as the number of learning images for image learning.

The sentence summarizer 140 generates a summary sentence by applying a correlation-related feedback and extracting a sentence using a term sentence matrix in which the extended query and term weights are redefined. Specifically, the cosine similarity is confirmed between the extended query and the weighted term sentence matrix, and a summary sentence is generated by extracting the upper sentences with the highest cosine similarity.

In summary, the query-based document summarizing apparatus 10 according to an embodiment of the present invention performs a preprocessing step, a query expansion step, a term weighting property renaming step, and a sentence generation step by sentence extraction.

First, the query-based document summarizing apparatus 10 decomposes an input document into sentences in a preprocessing step and extracts terms to form a term sentence matrix.

Secondly, the query based document summary device 10 expands the input user's query using the correlational feedback in the query expansion step.

Third, the query-based document summarizing apparatus 10 calculates a term sentence matrix in which the weights of terms are redefined using non-negative matrix-determined semantic features in the weight calculation step.

Finally, the query-based document summarizing apparatus 10 checks the similarity between the extended query and the term sentence matrix to which the redefined term weight is applied, and uses the higher preset number of sentences having a higher similarity to generate a summary sentence. Create

While the present invention has been described in detail with reference to the accompanying drawings, it is to be understood that the invention is not limited to the above-described embodiments. Those skilled in the art will appreciate that various modifications, Of course, this is possible. Accordingly, the scope of protection of the present invention should not be limited to the above-described embodiments, but should be determined by the description of the following claims.

Claims (2)

A preprocessor for decomposing an input document into sentences to generate a term sentence matrix;
A query extender that extends user queries using correlation feedback;
A weight calculator for recalculating the weights of terms using nonnegative matrixed semantic features; And
A summarizer that generates a summary for the input document using the extended terminology matrix with the user query and the weighted term of the redefined term
Query-based document summary device comprising a. The method of claim 1, wherein the summarizer,
And confirming the similarity between the term sentence matrix to which the weight of the redefined term is applied and the extended user query, and generating the summary by using the higher preset number of sentences having the high similarity.

Images