JPH07129605A - Document retrieval device - Google Patents

Abstract

PURPOSE:To understand the contents of a document by a user with the irreducible minimum labor and time by simultaneously displaying simple titles and summaries in the plural documents of retrieved results. CONSTITUTION:A system control part 19 detects retrieval instruction input to an input part 10 and starts a retrieval processing part 12 and the retrieval processing part 12 retrieves the corresponding document from a.= text data base 20 corresponding to a retrieval instruction. Then, the system control part 19 detects the completion of the processing of a format analysis part 13 and starts a summary generation part 14. The summary generation part 14 generates the summaries of the respective chapters of the document of the retrieved result and stores them in a summary sentence storage part 16. A retrieved result presenting part 17 makes the list of the retrieved results in a display means corresponding to summary sentences stored in the summary sentence storage part 16. A summary sentence presenting part 18 presents the summary sentence of the document selected in the retrieved result presenting part 17, also changes the detailed degree of the summary of a part corresponding to an instructed document position, inserts the summary sentence to the position and displays it.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document retrieval device for retrieving and presenting a desired document from a sentence database storing a plurality of documents.

[0002]

2. Description of the Related Art In recent years, the digitization of documents, such as the spread of word processors and computers, the spread of electronic news and electronic mail, and the publication of information (dictionaries, encyclopedias, etc.) in CDROM, which was previously only available on paper, has become possible. It's going on. In the future, it is expected that the digitization of such documents will accelerate.

Along with this, there is an increasing demand for an information retrieval system for retrieving a desired document from a large number of documents in order to effectively use the accumulated documents.

In the conventional document information retrieval system, the titles of the retrieval result documents are displayed side by side in a fixed order (for example, the order of documents similar to the keyword) according to the designation of a retrieval command such as a keyword. Met.
Therefore, it is difficult for the user to perceive the validity of the search result because the relative relationship between the searched documents and the degree of similarity with the search command cannot be realized by the user.

In the conventional document information retrieval system, the user must read the entire text of each document in detail in order for the user to determine whether or not the retrieved result is the desired document. Therefore, it took a lot of time to obtain a desired document.

By displaying only a part of the retrieved results, it is possible to reduce the labor required to read the whole sentence. However, if it is not clear how the displayed part relates to other parts of the document,
It cannot be determined whether the document is the desired document.
For example, if a part that describes background information about one document is displayed and a part that describes a conclusion about another document is displayed, which document is necessary because it cannot be viewed from a unified viewpoint. It is difficult to determine if Therefore, in order to understand how the displayed part is positioned with respect to the whole sentence, the whole sentence must be read, and as a result, the labor is not reduced.

Further, it is possible to manually create a summary sentence of each stored document and store it in association with the original document. By displaying this summary sentence in the display of the search result, it is possible to reduce the user's time and effort for verification. However, it takes a lot of time to manually create a summary sentence, and it is impossible to prepare unless the database is so large in scale. In addition, there are databases that do not yet have a summary of individual documents, and it is very costly and laborious to manually create a summary of them.

Further, since the summary text manually created is created for the general public, it is not always summarized from the viewpoint suitable for the purpose of the user. For this reason, there is a problem that the user may have a sentence that is out of focus, and may overlook a necessary document when determining whether or not the document is a desired document. In addition, the length of the manually created summary text cannot be changed according to the needs of the user. Therefore, while the user is reading the summary text, even if he / she wants to read that part in more detail, he / she cannot present the corresponding original text or a more detailed summary, and present the full text or print it out. It is necessary to read the entire text by doing so, and in such a case, there is a problem that the time and effort to read the searched document cannot be reduced.

For example, in PATOLIS, a plurality of summaries of search results can be listed at the same time, but if the user wants to know more detailed contents of the search results, the document must be newly printed out. As a result, the degree to which the effort involved in the search act was reduced was low.

[0010]

SUMMARY OF THE INVENTION The present invention has been made in consideration of the above-mentioned circumstances, and it is possible to interactively present a summary sentence whose detail level is changed according to a user's request. It is an object of the present invention to provide a document retrieval device that can be used.

[0011]

In the document retrieval apparatus of the present invention, a format analysis means for analyzing the structure of chapters from the title information of a document, and a variable length for the text of each chapter obtained by the format analysis means. , A search result presenting means for simultaneously presenting summaries of a plurality of documents obtained by the search results, and one of the summary texts presented by the search result presenting means. In addition to presenting a more detailed summary sentence or original document as a trigger, there is a means to specify the title, paragraph, sentence of the chapter or section on the presented summary sentence, and the summary sentence corresponding to the specified part Alternatively, it has a summary sentence presenting means for inserting the original sentence into the summary sentence being displayed.

The format analysis means analyzes the chapter structure of the original document from the relationship between the titles and headings assigned to each chapter and section in the document, the line feed and the appearance of white space characters.

The summary generating means generates a summary sentence of the sentences included in each chapter, based on the chapter structure obtained by the format analyzing means. In this summary generation, the length of the generated summary sentence can be made variable.

The search result presenting means has a function of presenting the title, author name, and abstract of each of the plurality of documents obtained as the search results. Further, it has a function of selecting one of a plurality of presented documents, and activates the summary document presenting means to present a more detailed summary sentence of the selected document.

The summary document presenting means presents at least a summary sentence that is more detailed than the summary sentence presented by the search result presenting section, and specifies a heading, a paragraph, or a part of the sentence in the presented summary sentence. Have. When the user makes this designation as appropriate, a more detailed summary (including the original text) of the corresponding part is inserted into the presented summary text.

[0016]

Since the simple title and the abstract are simultaneously displayed on a plurality of documents as the retrieval result, the trouble of having to read even a document that does not need to be read is greatly reduced. Also, if the detail of the presented summary is not sufficient, it is possible to dynamically specify that the corresponding part should be displayed in detail, so the content of the document obtained from the search results can be displayed to the user with the minimum required effort. Will be understandable. This makes it possible to improve the efficiency of search operations.

[0017]

Embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of equipment necessary for constructing the embodiment. Central processing means 1, storage means 2 such as semiconductor memory, magnetic disk, optical disk, display means 4 such as display such as CRT display, liquid crystal display, plasma display, display controller 3 for controlling display, keyboard, mouse, pen, etc. The input means 6 and the input controller 5 for controlling the input.

FIG. 2 is a block diagram showing the functional configuration of the embodiment. Input unit 11, search processing unit 12, format analysis unit 13, summary generation unit 14, document structure storage unit 15, summary text storage unit 16, search result presentation unit 17, summary text presentation unit 18, system control for controlling the overall system It consists of part 19. Text data to be searched is stored in the text database 20.

FIG. 3 illustrates a flow of processing of the system control unit 19 which controls the entire document retrieval apparatus of the embodiment.

The input unit 11 receives input of a keyword as a search command and a natural language sentence.

First, the system control unit 19 detects the input of a search command to the input unit and activates the search processing unit 12. The search processing unit 12 searches the text database 20 for the corresponding document according to the search command (step 31).

Next, the system control unit 19 detects the end of the processing of the search processing unit 12 and activates the format analysis unit 13 (step 32). The format analysis unit 13 analyzes the chapter structure of each document as a search result and stores it in the document structure storage unit 15.

Next, the system control section 19 detects the end of the processing of the format analysis section 13 and activates the summary generation section 14. The summary generation unit 14 generates a summary of each chapter of the search result document and stores it in the summary sentence storage unit 16 (step 3).
3).

Next, the system control unit 19 detects the end of the processing of the abstract generation unit 14 and activates the search result presentation unit 17 (step 34). The search result presentation unit 17 lists the search results on the display unit 4 shown in FIG. 1 according to the document structure of each document of the search results stored in the document structure storage unit 15 and the summary text stored in the summary document storage unit 16. Up (step 35).

From the displayed list of search results, 1
When one document is selected, the system control unit 19 activates the summary sentence presentation unit 18. The summary sentence presenting section 18
The summary text of the document selected by the search result presentation unit 17 is presented, the detail level of the summary of the portion corresponding to the document position instructed according to the user's instruction is changed, and the summary is inserted and displayed.

The search processing unit 12 is realized not by the gist of the invention of the present application but by using a method already implemented in a full-text search system or the like.

4 to 6 show the flow of processing of the format analysis unit 13.

The processing analyzes based on the title detection pattern. An example of the detection pattern is shown in FIG. The pattern is
A pattern for detecting chapter titles of a document as level 0, a pattern for detecting section titles constituting a chapter as level 1, and a pattern for detecting section titles at lower levels are set in advance.

In FIG. 7, the symbol "^" indicates the beginning of the line,
"Snumber" means to match a number, "Sstring" to an arbitrary character string, and "\ n" to a line break, respectively. For example, the level 0 detection pattern “^ [Snumber
] Sstring \ n "is" [1] Introduction "at the beginning of the line
If the string is present, the match with that line is successful. The analysis result is stored in the document structure storage unit 15 shown in FIG. 2 in the format shown in FIG. 8, for example.

In the format analysis unit 13, first, the pointer to the head address of the buffer in which the document is stored is set to P, the number of characters is set to N, and the detection level of the title is set to O to start the processing. The chapter counter is set to O, and P and N are stored in the title pointer and the number of characters of the document structure storage unit 15, respectively.

The position where the character string to be matched with the detection pattern exists is searched for. First, the start address of the character string matched with the detected pattern is stored in the body pointer (chapter counter + 1) as the body pointer. Then, the value of the chapter counter is incremented by 1. By sequentially repeating this process, the start addresses of the chapters are stored in the body pointer 1, the body pointer 2, and the body pointer 3, respectively. When the processing of the buffer in which the document is stored is completed for the number of characters N, the document structure storage unit 15 stores the number of chapters as a chapter counter, and the start address of each chapter is from body pointer 1 to body pointer n.
Stored in. Processing is performed from the first chapter pointed to by body pointer 1. The value of the body pointer T is stored in the title pointer T. Further, the number of characters is calculated from the difference from the body pointer T + 1 and is stored in the number of characters T. After that, the address of the title pointer T is stored in the body pointer T.
The value of the chapter counter T is set to O.

The position where the character string to be matched with the level 1 detection pattern exists is searched for. The start address of the character string to be matched with the detection pattern is stored in the body pointer (T, chapter counter + 1), and the chapter counter T is incremented by 1. A node structure is obtained by repeating this. When the value of the chapter counter T is 0 for the chapter T, it means that the clause does not exist. Therefore, in that case, the address of the character next to the chapter title is stored in the body pointer (T, 1). The chapter-clause structure is obtained by performing the above processing for all chapters.

FIG. 4 shows a flow in the case of analyzing a subordinate structure of a section (for example, analysis processing up to a level having a title of a chapter number such as "2.3.4").

The document structure analyzed by the format analysis unit 13 is stored in the document structure storage unit 15. For example, in the case of a document as shown in FIG. 9, as a result of the processing, the values shown in FIG. 10 are set in the document structure storage unit 15.

FIG. 11 shows the configuration of the summary generator 14.

The summary generation unit 14 in this embodiment comprises a context structure analysis unit 41, a penalty calculation unit 42, and a sentence reconstruction unit 43. The context structure analysis unit 41 extracts rhetorical expressions used in the document and analyzes the logical structure of the text portion of each chapter or section. Penalty calculation unit 42
Calculates the penalty for the degree of importance of each sentence based on the analyzed context structure.

The process itself in the context structure analysis unit 41 is
It is not the gist of the invention of the present application. For example, reference (A Discourse
Structure Analyzer for Japanese Text, K.Sumita, Int.
The method disclosed in Conf. On Fifth Generation Computer '92) can be used. By this process,
With respect to the sentence shown in FIG. 12B, the context structure shown in FIG. 12A can be obtained. The illustrated context structure shows that the second sentence has a relationship of “expansion” with respect to the first sentence, and the third sentence has a relationship of “planned” with respect to the first sentence and the second sentence. There is.

The context structure is stored in the summary sentence storage unit 15 as the data structure shown in FIG. The data structure represents the relationship (context structure.Rel) that the node has and the links (context structure.L, context structure.R) to the left and right nodes. However, in the case of a terminal node, the context structure. Rel is a pointer to the corresponding sentence, and context structure. L and context structure. -1 is stored in R.

FIG. 14 shows a processing flow of the penalty calculating section 42. The penalty calculation unit 42 advances the processing by recursively executing the penalty calculation. A terminal node with a small penalty at the end of processing becomes more important.

In the penalty calculation, the context structure analyzed by the context structure analysis unit 41 and the initial value of the penalty are input. First, it is determined whether the context structure is a terminal node. Context structure. If the value stored in L is −1, it is a terminal node, so the input penalty is the context structure. Set to P.

On the other hand, the context structure. If the value stored in L is not -1, it is not a terminal node but a non-terminal node. In this case, the context structure. It is determined whether the relation stored in Rel is the right-weighted type, the left-weighted type, or neither. This determination is performed based on the relationship table stored in advance. An example of the relationship table is shown below. Sequential right-weighted type Planned left-weighted type Illustrated left-weighted type The table above shows that the relationship "sequential" is the right-weighted type.

Context structure. If the relationship stored in Rel is right-weighted, the right node is more important, so
A value obtained by adding the penalty P1 to the penalty P is given to the penalty calculation on the left node side, and the penalty P is given to the penalty calculation on the right node side.

Conversely, the context structure. When the relation stored in Rel is a left-weighted type, the left node is important, so the penalty P is given to the penalty calculation on the side of the left node, and the value obtained by adding the penalty P1 to the penalty P is added to the right node. Give to the side penalty calculation.

Context structure. When the relation stored in Rel is neither right-weighted nor left-weighted, the penalty P is passed as it is to the penalty calculation for each of the left and right nodes.

By the above recursive processing, a small value is set for the penalty of a sentence of high importance, while a large value is set for the penalty of a sentence of low importance. Then, the penalty for each sentence is the summary document storage unit 1
Stored in 5.

FIG. 15 shows a processing flow of the text reconstructing section 43. The sentence reconstructing unit 43 refers to the penalties of all the sentences calculated by the penalty calculating unit 42, and takes out the sentences below the predetermined penalty. Then, the connection expression is changed to generate a summary sentence. For example, assume the following sentence: (Where 1, 2, 3 represent sentences). 1 “Good materials are needed to make good buildings.” 2 “But good tools are also needed.” 3 “In other words, materials are not enough.” And the context corresponding to this sentence The structure is as follows. (1 <reverse connection> (2 <duplication> 3)) Further, it is assumed that sentence 1 and sentence 3 are given a smaller penalty by the penalty calculation unit.

When sentence 1 and sentence 3 are simply connected, the following sentence which is logically different from the original sentence is generated. “Good materials are needed to make a good building.” “That is, materials are not enough.” Therefore, in changing the connection expression, the highest level relationship between sentences judged as key sentences was decided. Replace with the corresponding connection expression. In the above example,
Since the highest-level relationship between sentences 1 and 3 is <reverse connection>, the following sentence is generated by substituting the connection expression “but” corresponding to this relation. “To build a good building you need good materials.” “But
Only the materials are useless. By the sentence reconstructing section 43,
A summary sentence that is logically consistent with the original sentence is generated. Further, by changing the value of the penalty P2 in the sentence reconstructing unit 43, the amount of sentence generated as a summary sentence can be controlled. Therefore, it is possible for the user to control the amount of displayed summary by resetting the penalty P2.

FIG. 16 illustrates a processing flow of the search result presentation unit 17.

First, for each document in the search result,
The title and the author name are extracted from the document stored in the document structure storage unit 15. For example, for the document example shown in FIG. 9, the title “Automatic document layout system” and the author names “Taro Yamada, Hanako Sato” are extracted.

Next, the context structure corresponding to the "outline" portion is extracted from the summary sentence storage unit 16 in correspondence with the above steps. Then, the sentence reconstructing process 43 shown in FIG. 15 is activated to generate a summary sentence of the "outline" portion. And
A list of search results as shown in FIG. 17 is displayed on the display unit 4 shown in FIG. For example, for the document shown in FIG. 9, the display shown at the top is performed as follows. "Automatic layout system for sentences Taro Yamada, Hanako Sato This article describes the technical details." After displaying, the search result presentation unit 17 waits for an input event from a pointing device such as a mouse. When the user selects one of the listed documents with the mouse, the summary sentence presenting unit 18 is activated for the selected document. FIG. 18 illustrates a processing flow of the summary text presentation unit 18. 19 to 22 show display examples in the summary sentence presenting section 18.

First, the summary sentence presentation unit 18 displays the summary sentence of the document selected by the search result presentation unit 17. In this display, the title, the author name, and the outline portion are displayed as in the search result presentation unit 17, and the heading of each chapter of the document is also displayed by referring to the document structure storage unit 15. FIG. 19 shows a display example when the first document is selected from the list shown in FIG.

Next, the display is input by the user,
If the command is a detail level change command, the chapter number is obtained from the designated position and the detail level input menu is displayed as shown in FIG. Then, when the user inputs the degree of detail, the penalty P2 when the sentence reconstructing unit 43 generates the sentence is determined by the following formula, and the summary sentence of the corresponding chapter is generated. P2 = C × (detail level) Here, C is a predetermined constant.

Next, the summary sentence presenting section 16 inserts the newly generated summary sentence into the initially displayed summary sentence and displays it. This state is shown in FIG. The hatched portion in FIG. 21 is the additional portion of the newly generated summary sentence.

In FIG. 21, the outline portion is designated and the summary text of the portion is changed and displayed. However, when the heading portion of each chapter is designated as shown in FIG. 22,
A new summary of the specified chapter will be displayed.

The summary sentence presenting section 18 repeats the above processing. This allows the user to change the level of detail of the summary text of the desired chapter. When the detail level is 100%, the original text is displayed.

The summary generating unit 14 of the embodiment generates the summary sentence by analyzing the context structure. However, other conventionally disclosed methods (for example, the frequency of a used word is calculated to determine the frequency). It is easy to transform it into a method that takes out a sentence containing a high word or a method that simply cuts out the sentence from the front of the sentence).

Further, the search result presentation unit 17 of the embodiment directly displays the summary sentence generated by the sentence reorganization unit 43. Therefore, as shown in FIG. 17, the summary sentence is displayed over three lines for the third document. However, it is also possible to limit the portion of the summary text to a certain length and present it as shown below. "DSI's CAD, Kazuo Tanaka, among various kinds of LSI's CAD, as a representative of EA, custom VLSI CAD and analog layout CAD" In the summary sentence presenting section 18 of the embodiment, at the time of start-up, the title, author's name, and summary portion are summarized. Although the sentences and the headings of the respective chapters are displayed, it is easy to transform them so as to present, for example, a summary document about the postscript part. Similarly, it is possible to allow the user to customize the display in various forms. Further, in the embodiment, the detail level is set one by one as shown in FIG. 20, but it is possible to omit the setting level by increasing the detail level at a constant rate. Is.

The search processing unit 12 of the embodiment performs a search process for a local text database that is in the search device, but it is easy to change this to a database that can be accessed online. .

Although a plurality of abstract sentences are presented at the same time, they may be presented one after another or may be presented by dividing them by a predetermined number. Furthermore, the degree of detail corresponding to the user's request described above can be designated from the outside.

Next, another embodiment will be described with reference to FIGS.

In this embodiment, the summary document presenting section 18 of FIG. 2 is used.
Except for the processing contents of the above, it is the same as the previous embodiment, and the detailed description of the same parts will be omitted.

FIG. 23 shows a processing flow of the summary document presenting section 18 according to this embodiment. According to the flow of this process, at the start, first, titles of chapters or sections from level 0 to level M are displayed as shown in FIG. Then wait for an input event. When there is an input, it is determined whether or not the input is a detail level change command.
If this determination is YES, it proceeds to the step of determining whether the level M is the maximum value of the level, and if NO, it proceeds to the step of determining whether it is an end command.

If the determination result of the level M is NO, M is incremented by 1, the process returns to the first step, and YES.
Then, the level L and chapter number Q of the chapter or section including the position of the cursor are obtained. After that, the detail level input menu is displayed to wait for the input of the detail level. When the detail level command is designated, the section headings are additionally displayed as shown in FIG. Furthermore, as shown in FIG.
The headings (1) and (2) in section 3.2 are displayed.

Further, when the detail level command is specified,
The abstract text of section 2.1 designated by the cursor is displayed as shown in FIG. At this time, if the cursor is placed at the position of the chapter 2 heading,
As shown in FIG. 26, the abstract sentences of the sentences in sections 2.1 and 2.2 are displayed immediately after the chapter 2 heading.

Next, the document reconstruction process is performed according to the input degree of detail. The summary document generated by this document reconstruction process is inserted into the text being displayed. After that, the chapter number Q is incremented, and it is determined whether or not the level of the chapter or section of the chapter number is smaller than L. If this determination is YES, the process returns to the step of document reconstruction processing, and N
If it is O, the process returns to the step of waiting for an input event.

If the judgment is NO in the above-mentioned judgment step for judging whether or not the input is the detail change command,
Whether the input is an end command is determined. If the input command is an end command, the window is erased and the operation ends. If the determination is NO, the process returns to the step of waiting for an input event.

As described above, according to the above embodiment, the abstract text is displayed according to the specification of the detail command,
Since the summary document is created by performing the document reconstruction process according to the input level of detail and this summary document is applied to the search, the content of the search target sentence is used with the minimum required effort in the document search. Can be understood by others.

The present invention is not limited to the above embodiments, and various modifications can be made without departing from the spirit of the invention.

[0070]

As described above, in the document search device of the present invention, it is possible to dynamically present a specified chapter or section individually as a summary sentence having a different level of detail. As a result, the user can understand the content written in each document of the search result with the minimum required effort, and an efficient search action can be realized.

[Brief description of drawings]

FIG. 1 is a device configuration diagram.

FIG. 2 is a functional configuration diagram.

FIG. 3 is a diagram showing a processing flow of a system control unit.

FIG. 4 is a diagram showing a processing flow of a format analysis unit.

FIG. 5 is a diagram showing a processing flow of a format analysis unit.

FIG. 6 is a diagram showing a processing flow of a format analysis unit.

FIG. 7 is a diagram showing a title detection pattern.

FIG. 8 is a diagram showing a format of a document structure recording unit.

FIG. 9 is a diagram showing an example of a document.

FIG. 10 is a diagram showing an example of contents of a document structure recording unit.

FIG. 11 is a diagram showing a flow of processing of a summary generation unit.

FIG. 12 is a diagram showing an example of a document and a context structure.

FIG. 13 is a diagram showing an example of a context structure stored in a summary text storage unit.

FIG. 14 is a diagram showing a processing flow of a penalty calculation unit.

FIG. 15 is a diagram showing a flow of processing of a sentence reconstructing unit.

FIG. 16 is a diagram showing a processing flow of a search result presentation unit.

FIG. 17 is a diagram showing a display example in the search result presentation unit.

FIG. 18 is a diagram showing a flow of processing of a summary text presentation unit.

FIG. 19 is a diagram showing a display example in a summary text presentation unit.

FIG. 20 is a diagram showing a display example in a summary text presentation unit.

FIG. 21 is a diagram showing a display example in a summary text presentation unit.

FIG. 22 is a diagram showing a display example in a summary text presentation unit.

FIG. 23 is a flowchart showing the processing contents of a summary document presenting unit according to another embodiment.

FIG. 24 is a diagram showing an example of a display in a summary sentence presentation unit based on the embodiment of FIG. 24.

FIG. 25 is a diagram showing another example of the display in the summary sentence presentation unit based on the embodiment of FIG. 24.

FIG. 26 is a diagram showing another example of the display in the summary sentence presentation unit based on the embodiment of FIG. 24.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Central processing means 2 ... Storage means 3 ... Display controller 4 ... Display means 5 ... Input controller 6 ... Input means

Claims (1)

[Claims] 1. Storage means for storing a plurality of documents, search means for searching the documents stored in the storage means, and analysis means for analyzing the structure of the document obtained by the search means. Summary generating means for generating a summary sentence based on the document structure obtained by the analyzing means, and a search result for presenting simultaneously or sequentially a plurality of summary sentences for the document obtained by the summary generating means The presenting means, the selecting means for selecting one of the summary sentences presented by this search result presenting means, and the more detailed summary sentence or the original document itself is presented for the summary sentence selected by this selecting means. A document retrieval device characterized by the above.