JP2005242807A - Related knowledge retrieval apparatus, sentences network generation device, sentences network generation method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To construct a sentences network without being influenced by a personal point of view and to retrieve flexibly desired related knowledge that has arbitrary order with respect to knowledge to be inspected. <P>SOLUTION: By a sentences network generation means 15D, sentences including the same important word as a starting point sentence designated by teaching information 14B is selected as a starting point candidate sentence from among a sentences set 14A, and sentences including the same important word as an end point sentence designated by the teaching information 14B is selected as an end point candidate sentence from among the sentences set 14A. By using new directed line segment with the starting point sentence and the starting point candidate sentence as starting points respectively and with the end point sentence and the end point candidate sentence as an end point respectively, a sentences network to indicate the relation and order among the respective sentences is generated. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a related knowledge search technique, and more particularly to a related knowledge search technique for searching a document having knowledge related to desired knowledge from a plurality of documents.

  In an advanced information society, with the development of information processing technology and information communication technology, an enormous amount of digitized information is easily stored and an environment for using it is being provided. In order to solve an arbitrary problem using such an environment, it is necessary to efficiently search knowledge related to the problem from a huge amount of information.

Conventionally, as a technique for searching for knowledge related to such an arbitrary problem, a technique for searching based on the similarity of each sentence has been proposed (see, for example, Patent Document 1).
In this related knowledge search technology, in order to efficiently determine the similarity of a sentence (text data) indicating the contents of an event with other sentences, the sentence structure and the The similarity regarding the meaning is calculated | required, both are collated and the sentence with the high similarity is displayed and output as a similar sentence similar to the sentence of an input event.

However, according to the related knowledge search technique, there is a problem that only similar sentences having high similarity to the sentence of the input event are searched and displayed, and the order between the sentences cannot be obtained. . For example, when searching for an input event that occurred in equipment maintenance / operation work, only an event similar to that input event is presented, and related events such as the event that caused the event and the event that can occur afterwards Can't get about.
In the maintenance and operation of facilities, as described above, it is necessary to take prompt and appropriate actions for the events that have occurred. To that end, the causes of input events and the events that can occur afterwards are related to the input events. It is important to accurately grasp the relevant circumstances.

On the other hand, an expert who is familiar with the contents of each sentence to be classified teaches in advance the order and relevance between documents, and constructs a sentence network that shows the order and relevance between sentences in advance. In addition, based on the sentence network, a method in which an unskilled person searches for a sentence related to an arbitrary sentence can be considered.
According to this method, not only the relationship between each sentence but also the order between documents can be obtained, so it is possible to easily grasp the cause of the input event and the event that can occur afterwards. It can also be used for operations.

The applicant has not yet found prior art documents related to the present invention by the time of filing other than the prior art documents specified by the prior art document information described in this specification.
JP 2000-276487 A

However, since the conventional technique uses a sentence network constructed based on the subjectivity of the expert's teaching, the order and relationship between sentences greatly depends on the instruction of the expert, There was a problem that related knowledge could not be searched from a new point of view beyond the scope of teaching.
The present invention is to solve such a problem, and it is possible to construct a sentence network without being constrained by subjectivity and to flexibly search for desired related knowledge having an arbitrary order with respect to knowledge to be examined. It is an object of the present invention to provide a search device, a text network generation device, a text network generation method, and a program.

  In order to achieve such an object, the related knowledge search device according to the present invention includes a sentence set composed of a plurality of sentences in which various pieces of knowledge are described as character information using a predetermined important word, and each sentence belonging to the sentence set. A storage unit for storing sentence network information indicating a relation between sentences, and a calculation process for searching a desired sentence related to a sentence including knowledge to be searched from a sentence set in the storage unit, using the sentence network information in the storage part. A related knowledge search device having an input to the arithmetic processing unit, the inter-sentence relation information indicating the relation and order between the two sentences, and the start point located on the start point side of the inter-sentence relation information Sentence network generation means for generating sentence network information is provided based on important words contained in the sentence and the end point sentence located on the end point side of the inter-text relation information. The text generator selects a sentence including all the same important words as the starting sentence specified by the input sentence relation information from the sentence set in the storage unit as a starting point candidate sentence, and an end point specified by the sentence relation information. Select a sentence that contains all the same important words as the sentence from the set of sentences in the storage section as the end point candidate sentence, and start a new sentence between each of the start point sentence and the start point candidate sentence, and the end point sentence and the end point candidate sentence. Generation of relationship information, and sentence network information indicating the relationship and order between each sentence from the pair of information indicating the start sentence and the information indicating the end sentence of the new inter-sentence relation information. Is.

  At this time, in the arithmetic processing unit, arbitrary sentence relation information, each of the important words included in the end sentence of the inter-sentence relation information starting from each of the important words included in the start sentence of the inter-sentence relation information. A sentence weight calculation means for decomposing the important word relation information as the end point and calculating the sentence weight corresponding to the sentence relation information based on the sum of the important word weights corresponding to the important word relation information. Furthermore, a sentence network having a sentence weight obtained by the sentence weight calculating means may be generated for each sentence by the sentence network generating means.

  Further, the arithmetic processing unit is configured to input the inter-sentence relationship information, and each of the important words included in the end-point sentence of the inter-sentence relation information starting from each of the important words included in the start-point sentence of the inter-sentence relation information. There may be further provided an important word weight calculation means for decomposing each important word relation information as the end point and updating the important word weight corresponding to the important word relation information for each occurrence of the important word relation information. Good.

  In addition, the arithmetic processing unit further includes a knowledge search unit that searches the sentence set in the storage unit for a sentence having a sentence including a desired knowledge as a start point sentence or an end point sentence based on the sentence network information, and outputs the sentence as a sentence related to the knowledge. It may be provided.

  In addition, the arithmetic processing unit searches the sentence set of the storage unit for a sentence having a sentence including the desired knowledge based on the sentence network information as a start point sentence or an end point sentence, and obtains the obtained sentence as knowledge in the order of the weight between the sentences. You may further provide the knowledge search means output as a related sentence.

  The sentence network information generating apparatus according to the present invention relates to a storage unit that stores a sentence set composed of a plurality of sentences in which various kinds of knowledge are described using character information using a predetermined important word, and a sentence that includes search target knowledge. A sentence network information generation device having a calculation processing unit that generates a sentence network information used for searching a desired sentence to be searched from a set of sentences in a storage unit, the relevance between two sentences in the calculation processing unit And the inter-sentence relationship information indicating the order, and based on the important words included in the start point sentence located on the start point side of the inter-sentence relationship information and the end point sentence located on the end point side of the inter-sentence relationship information, Sentence network generation means for generating sentence network information is provided, and this sentence network generation means uses the same important word as the start sentence specified by the input sentence relationship information. Select all the sentences including all as the start point candidate sentences from the sentence set in the storage unit, and select all sentences including all the same important words as the end point sentence specified in the inter-sentence relationship information from the sentence set in the storage unit as the end point candidate sentences, Generate new inter-sentence relationship information with each of the start point sentence and start point candidate sentence as the start point and each of the end point sentence and end point candidate sentence as the end point. You may make it produce | generate the text network information which shows the relationship and order between each text from the pair with the information to show.

  The sentence network generation method according to the present invention relates to a storage unit that stores a sentence set composed of a plurality of sentences in which various kinds of knowledge are described as character information using a predetermined important word, and a sentence including knowledge to be searched. Sentence network information based on input predetermined sentence relationship information used in a sentence network information generation apparatus having an arithmetic processing unit for generating sentence network information used to search a desired sentence from a sentence set in a storage unit A sentence network generation method for generating a sentence network, wherein the arithmetic processing unit receives inter-sentence relation information indicating relevance and order between two sentences, and is a start-point sentence located on the start-point side of the inter-sentence relation information Network generation that generates text network information based on key words contained in the end-point text located on the end-point side of the inter-text relation information In this sentence network generation step, a sentence including all the same important words as the starting sentence specified by the inputted sentence relation information is selected as a starting point candidate sentence from the sentence set in the storage unit, and the sentence relation Selecting a sentence including all the same important words as the end sentence specified in the information as an end point candidate sentence from the sentence set in the storage unit, and each of the end point sentence and the end point candidate sentence from the start point sentence and the start point candidate sentence respectively. Is used to generate new inter-sentence relationship information, and from these pairs of information indicating the start sentence of the new inter-text relation information and information indicating the end sentence, a sentence indicating the relationship and order between the sentences. Generating network information.

  At this time, in the arithmetic processing unit, arbitrary sentence relation information is obtained by using each important word contained in the end sentence of the sentence relation information starting from each of the important words contained in the start sentence of the sentence relation information. A sentence weight calculation step for calculating a sentence weight corresponding to the sentence relation information based on the sum of the weights of the important words corresponding to the important word relation information. Further, a sentence network having a sentence weight obtained in the sentence weight calculation step may be generated for each sentence in the sentence network information generation step.

  Further, in the arithmetic processing unit, the input inter-sentence relation information is obtained by using each of the important words included in the end sentence of the inter-sentence relation information as the start point of each of the important words contained in the start sentence of the inter-sentence relation information. There may be further provided an important word weight calculation step for decomposing each important word relation information as the end point and updating the important word weight corresponding to the important word relation information for each occurrence of the important word relation information. Good.

  In addition, the program according to the present invention includes a storage unit that stores a sentence set including a plurality of sentences in which various kinds of knowledge are described using character information using predetermined important words, and a desired sentence related to the sentence including the search target knowledge. A sentence network information generating apparatus having an arithmetic processing unit for generating sentence network information used to retrieve a sentence set from a sentence set in a storage unit, and inter-sentence relation information indicating relevance and order between two sentences Is selected as a starting point candidate sentence from the sentence set in the storage unit, and located on the end side of the inter-sentence relation information. Selecting a sentence including the same important word as the end sentence from the sentence set in the storage unit as an end point candidate sentence, and setting each of the start point sentence and the start point candidate sentence as a start point Generate new inter-sentence relationship information with the end point sentence and the end point candidate sentence as the end point, and from the pair of information indicating the start point sentence and the end point sentence of these new inter-sentence relation information, Generating sentence network information indicating relevance and order.

  According to the present invention, a sentence including the same important word as the start sentence specified by the teaching information is selected as a starting point candidate sentence from the sentence set, and a sentence including the same important word as the end sentence specified by the teaching information is selected. Using the new directed line segment selected from the sentence set as the end point candidate sentence and starting with each of the start point sentence and the start point candidate sentence and ending with the end point sentence and each of the end point candidate sentences, the relationship between the sentences and Sentence network that shows order is generated, it is possible to build a sentence network that is not bound by the subjectivity of expert instruction, and flexibly add desired related knowledge with arbitrary order to search target knowledge Searchable.

Next, embodiments of the present invention will be described with reference to the drawings.
First, a related knowledge search device according to an embodiment of the present invention will be described with reference to FIG. FIG. 1 is a block diagram showing a configuration of a related knowledge retrieval apparatus according to an embodiment of the present invention.

The related knowledge search device 1 according to the present embodiment, as a sentence network used for related knowledge search, for a start sentence and an end sentence specified by teaching information, a sentence including the same important word as these sentences from a sentence set as a start point candidate Relevance and ordering between sentences based on new sentence relationship information that is selected as a sentence and an end point candidate sentence, and each of the start point sentence and the start point candidate sentence is the start point and the end point sentence and the end point candidate sentence are the end points. A network of sentences is generated.
In the following, the inter-sentence relationship information indicating the relationship between the two sentences and the important word relationship information indicating the relationship between the two important words are each represented by a directed line segment based on the graph theory.

The related knowledge search device 1 is composed of an information processing device having a computer as a whole, and is provided with a screen display unit 11, an operation input unit 12, an information input / output unit 13, a storage unit 14, and an arithmetic processing unit 15.
The screen display unit 11 includes a screen display device such as an LCD or a CRT, and displays various types of information on the screen in accordance with the output from the arithmetic processing unit 15.
The operation input unit 12 includes an input device such as a keyboard and a mouse, detects a user operation, and outputs the detected operation to the arithmetic processing unit 15.
The information input / output unit 13 is a circuit unit that is connected to a recording medium or a communication network and exchanges various processing information and programs used for processing in the arithmetic processing unit 15.

The storage unit 14 includes a storage device such as a hard disk or a memory, and stores various processing information 14A to 14E and a program 14F used for processing in the arithmetic processing unit 15.
As processing information, a sentence set 14A composed of a plurality of sentences in which various kinds of knowledge are described in text data, teaching information 14B indicating a relationship between arbitrary sentences taught by an expert, and sentences in the sentence set 14A are analyzed. An important word list 14C composed of a plurality of important words for grasping the contents, sentence network information 14D indicating a relation between sentences in the sentence set 14A, and a relation between important words used in each sentence in the sentence set 14A. There is important word network information 14E.

The arithmetic processing unit 15 is composed of a CPU and its peripheral circuits, and reads and executes the program 14F in the storage unit 14, thereby realizing various functional means by cooperating the hardware and the program.
As functional means, there are teaching information reception means 15A, sentence network generation means 15B, sentence weight calculation means 15C, keyword network generation means 15D, keyword weight calculation means 15E, and knowledge search means 15F.

The teaching information receiving unit 15 </ b> A is a functional unit that receives an operation from the operation input unit 12, generates teaching information 14 </ b> B indicating a relationship between arbitrary sentences taught by a skilled person, and stores it in the storage unit 14.
The sentence network generation unit 15B is a functional unit that analyzes the relationship between each sentence in the sentence set 14A based on the teaching information 14B and the important word list 14C, and generates the sentence network information 14D.
The sentence weight calculation means 15C is a functional means for calculating a weight between sentences constituting the sentence network information 14D based on the number of important words included in two sentences.

The keyword network generation unit 15D is a functional unit that analyzes the relationship between the key words included in each sentence of the sentence set 14A based on the teaching information 14B and the keyword list 14C, and generates the keyword network information 14E.
The important word weight calculation means 15E is a functional means for calculating the weight between the important words constituting the important word network information 14E based on the teaching information 14B.
The knowledge search unit 15F is a functional unit that searches for a sentence related to a desired sentence based on the sentence network information 14D and outputs it as related knowledge.

FIG. 2 shows a configuration example of the sentence set 14A. In this example, “Stress” is a summary of sentences that a number of respondents freely described on the Web, and a sentence number Di for managing the sentence D for each sentence D Is assigned.
FIG. 3 shows a configuration example of the important word list 14C. This important word list 14C is constructed by analyzing each sentence D based on a predetermined algorithm and constructing each important word T from the obtained important word type and its context, and for each important word T, A term number Tj for managing the important word T is assigned.

Each important word T consists of a pair of a front keyword and a rear keyword after the keyword. A word indicating the content of the keyword and a part of speech attribute type of the word are defined for each keyword. .
For example, the important word “1” is composed of two keywords “stress” and “elimination”, and the positional relationship is defined such that “stress” is located in front. The important word “2” is also composed of two keywords “stress” and “elimination”, and the positional relationship is defined such that “stress” is located behind.

[Text network generation operation]
Next, with reference to FIG. 4, the sentence network generation operation in the related knowledge search device according to the first exemplary embodiment of the present invention will be described. FIG. 4 is a flowchart showing a sentence network generation process in the related knowledge search device according to the first embodiment.
In response to a predetermined sentence network generation instruction operation from the operation input unit 12 or completion of reception of new teaching information 14B in the teaching information receiving unit 15A, the arithmetic processing unit 15 uses the sentence network generating unit 15B to perform the operation shown in FIG. Start the sentence network generation process.

  First, the sentence network generation means 15B, for each sentence included in the sentence set 14A of the storage unit 14, for the start point sentence Ds located on the start point side of the directed line segment of the teaching information 14B and the sentence, the inter-sentence weight The calculation means 15C calculates the sentence weight v (step 100). At this time, the weight between the sentences is calculated with the starting point sentence Ds as the starting point side and each sentence of the sentence set 14A as the ending point side.

  The sentence weight v is obtained by the following expression (1) by the sentence weight calculation means 15C. In Expression (1), | De | indicates the number of important words included in the end point sentence De located on the end point side of the directed line segment of the teaching information 14B, and | De∩DOCd | is calculated as the end point sentence De. The number of important words included in common with the sentence DOCd located on the end point side of the directed directed line segment is shown.

  Therefore, according to the formula (1), when all the important words included in the end point sentence De are included in the sentence DOCd, the variable c = 1 is defined, and the inter-sentence weight v is also “1”. . On the other hand, when all of the important words included in the end point sentence De are not included in the sentence DOCd, the variable c = 0 is defined, and the inter-sentence weight v is also “0”.

  In this way, the inter-sentence weight v between the starting point sentence Ds and each sentence of the sentence set 14A is calculated, and the sentence where the inter-sentence weight v is an effective value, that is, a value other than zero, is determined as the end point with respect to the starting point sentence Ds. A candidate sentence is selected (step 101), and a new directed line segment is generated for the end point sentence De and each end point candidate sentence from the start point sentence Ds (step 102).

Next, for each sentence included in the sentence set 14 </ b> A of the storage unit 14, the sentence network generation unit 15 </ b> B calculates the sentence interval between the sentence and the end point sentence De located on the end point side of the directed line segment of the teaching information 14 </ b> B. The weight calculation means 15C calculates the sentence weight v (step 103). At this time, each sentence in the sentence set 14A is set as the start point side and the start point sentence Ds is set as the end point side, and the weight between the sentences is calculated. Further, in Expression (1), Ds may be used instead of De, and text DOCs positioned on the starting point side of the directed line segment to be calculated may be used instead of DOCd.
Then, a sentence whose inter-sentence weight v is an effective value, that is, a value other than zero is selected as a starting point candidate sentence for the end point sentence De (step 104), and from each start point candidate sentence to the end point sentence De and the end point candidate sentence. A new directed line segment is generated (step 105).

As a result, a sentence including all the important words included in the end point sentence De is selected as an end point candidate sentence from the sentence set 14A, and a sentence including all the important words included in the start point sentence Ds is selected from the sentence set 14A. Selected as a candidate sentence, new directed line segments are generated, each having a start point sentence Ds and a start point candidate sentence as a start point, and an end point sentence De and an end point candidate sentence as an end point.
Then, a pair of the start point sentence and the end point sentence indicating the directed line segment is output as the sentence network information 14D and stored in the storage unit 14 (step 106), and a series of sentence network generation processes is completed.

FIG. 5 shows an example of a sentence network generation operation. In this example, as shown in FIG. 5A, it is assumed that the directed line segment Vk0 having the sentence D163 as the start point sentence Ds and the sentence D357 as the start point sentence De is specified as the teaching information 14B.
Then, in steps 100 and 101 of FIG. 4, from the sentence set 14A, as shown in FIG. 5B, the end points at which the inter-sentence weights v1 and v2 with the start sentence D163 become valid values (other than zero). Candidate sentences D380 and D421 are selected, and in step 102 in FIG. 4, directed line segments Vd1 and Vd2 are generated on the sentence network from the start point sentence D163 to the end point candidate sentences.

Subsequently, in step 103 of FIG. 4, a starting point candidate sentence D196 in which the inter-sentence weight v3 with the end point sentence D357 is an effective value is selected from the sentence set 14A, as shown in FIG. 5C. In step 104 of FIG. 4, directed line segments Vd3 to Vd5 are generated from the start point candidate sentence D196 to the end point sentence D357 and the end point candidates D380 and D421, respectively.
These directed line segments Vk0, Vd1 to Vd5 are output as the sentence network information 14D. FIG. 6 shows a configuration example of the text network information 14D. Here, the sentence number of the start sentence located on the start point side of each directed line segment and the sentence number of the end sentence located on the end point side are described as a pair.

  As described above, for the start point sentence Ds and the end point sentence De specified by the teaching information 14B, sentences including the same important words as these sentences are selected as the start point candidate sentence and the end point candidate sentence from the sentence set, and the start point sentence Ds and the start point candidate are selected. Based on new directed line segments with each sentence as the start point and each of the end point sentence De and the end point candidate sentences as end points, a sentence network showing the relationship and order between each sentence is generated. The information 14B can be analyzed at the important word level, and a sentence network including a new directed line segment that is not specified by the expert's instruction can be easily constructed. Thereby, it is possible to flexibly search for related knowledge having order from a new viewpoint that exceeds the subjective range of teaching by an expert.

  Next, with reference to FIG. 7 and FIG. 8, the sentence network generation operation in the related knowledge search device according to the second exemplary embodiment of the present invention will be described. FIG. 7 is a flowchart illustrating the keyword network generation processing in the related knowledge search device according to the second embodiment. FIG. 8 is a flowchart showing a sentence network generation process in the related knowledge search device according to the second embodiment.

In the first embodiment described above, when a new directed line segment is generated in the sentence network, a sentence including all the same important words as the sentence for the start sentence and the end sentence specified by the teaching information The case of selecting as a candidate sentence and an end point candidate sentence has been described as an example.
In this embodiment, the directed line segment specified by the teaching information is decomposed into directed line segments between important words included in the start sentence and the end sentence, and weights between the important words are assigned. A case will be described in which important word network information composed of direction line segments is generated, and start point candidate sentences and end point candidate sentences used when generating new directional line segments are selected based on the important word network information.

In response to a predetermined sentence network generation instruction operation from the operation input unit 12 or completion of reception of new teaching information 14B in the teaching information receiving unit 15A, the arithmetic processing unit 15 first uses the important word network generating unit 15D to The important word network generation process of FIG. 7 is started.
The keyword network generation unit 15D converts the directional line segment Vk specified by the teaching information 14B of the storage unit 14 into the keyword and the end point sentence De located on the end point side included in the start point sentence Ds located on the start point side. The important words included are acquired (step 110).

  Then, by generating a directional line segment between the start point side important word and the end point side important word, the directional line segment specified by the teaching information 14B is decomposed into the important inter-word directional line segment (step). 111). At this time, each time the important inter-word directed line segment is generated, the important inter-word weight w corresponding to the important inter-word directed line segment is updated (step 112).

The important word weight w is obtained by the following important expression (2) by the important word weight calculating means 15E. In the equation (2), w _{0 is} an important word weight between the original words, and w ₀ = 0 in the initial state where the important inter-word directed line segment is not generated. α is a coefficient for determining the (include) ratio with respect to the original important word weight w ₀ , β is a coefficient for determining the + side update width with respect to the original important word weight w ₀ , and γ is the original importance A coefficient for determining a negative update width for the inter-word weight w ₀ is shown. The coefficients α, β, and γ all take values of 1 or more. The initial value of w ₀ (when a new important line between important words is generated) is set to zero.

In this manner, the important word weight w is calculated for each important word directed line segment. Then, these important inter-word directed line segments and the important inter-word weight w are output as the important word network information 14E and stored in the storage unit 14 (step 113), and a series of important word network generation processes is completed.
As a result, the important word network information is generated as shown in FIG. Here, the important word number of the starting important word located on the starting point side of each directed line segment, the important word number of the ending important word located on the ending point side, and the weight w between the important words are described as a pair. Yes.

Next, the arithmetic processing unit 15 starts the sentence network generation process of FIG. 9 using the sentence network generation unit 15B.
First, the sentence network generation means 15B, for each sentence included in the sentence set 14A of the storage unit 14, for the start point sentence Ds located on the start point side of the directed line segment of the teaching information 14B and the sentence, the inter-sentence weight The calculation means 15C calculates the sentence weight v (step 120). At this time, the weight between the sentences is calculated with the starting point sentence Ds as the starting point side and each sentence of the sentence set 14A as the ending point side.

  The sentence weight v is obtained by the following expression (3) by the sentence weight calculation means 15C. In Expression (3), | De | indicates the number of important words included in the end point sentence De located on the end side of the directed line segment of the teaching information 14B, and | De∩DOCd | is calculated as the end point sentence De. The number of important words included in common with the sentence DOCd located on the end point side of the directed directed line segment is shown. M represents the number of important words included in the sentence Ds located on the start point side of the directed line segment to be calculated, and N represents the sentence DOCd located on the end point side of the directed line segment to be calculated. Indicates the number of important words.

  Therefore, according to Equation (3), when all the important words included in the end point sentence De are included in the end point side sentence DOCd, the variable c = 1 is defined, and the important word and the end point of the start point sentence Ds are defined. The sum of the weights w of each important word directed line segment with the important word of the side sentence DOCd becomes the inter-sentence weight v. On the other hand, when all of the important words included in the end point sentence De are not included in the end point side sentence DOCd, the variable c = 0 is defined, and the inter-sentence weight v is also “0”.

  In this way, the inter-sentence weight v between the starting point sentence Ds and each sentence of the sentence set 14A is calculated, and the sentence where the inter-sentence weight v is an effective value, that is, a value other than zero, is determined as the end point with respect to the starting point sentence Ds. A candidate sentence is selected (step 121), and a new directed line segment is generated for the end point sentence De and each end point candidate sentence from the start point sentence Ds (step 122).

Next, for each sentence included in the sentence set 14 </ b> A of the storage unit 14, the sentence network generation unit 15 </ b> B calculates the sentence interval between the sentence and the end point sentence De located on the end point side of the directed line segment of the teaching information 14 </ b> B. The weight calculation means 15C calculates the sentence weight v (step 123). At this time, each sentence in the sentence set 14A is set as the start point side and the start point sentence Ds is set as the end point side, and the weight between the sentences is calculated. Further, in Expression (3), Ds may be used instead of De, and text DOCs positioned on the start point side of the directed line segment to be calculated may be used instead of DOCd.
Then, a sentence whose inter-sentence weight v is an effective value, that is, a value other than zero, is selected as a start point candidate sentence for the end point sentence De (step 124), and from each start point candidate sentence to the end point sentence De and the end point candidate sentence. A new directed line segment is generated (step 125), and an inter-text weight v is calculated for each of these directed line segments using the inter-text weight calculation means 15C (step 126).

As a result, a sentence including all the important words included in the end point sentence De is selected as an end point candidate sentence from the sentence set 14A, and a sentence including all the important words included in the start point sentence Ds is selected from the sentence set 14A. Selected as a candidate sentence, new directed line segments are generated, each having a start point sentence and a start point candidate sentence as a start point, and an end point sentence De and an end point candidate sentence as an end point.
Then, a pair of the start point sentence, the end point sentence indicating the directed line segment, and the weight between the sentences is output as the sentence network information 14D and stored in the storage unit 14 (step 127), and a series of sentence network generation processes is completed. To do.

  FIG. 10 shows a configuration example of the text network information 14D. Here, the sentence number of the start sentence located on the start point side of each directed line segment, the sentence number of the end sentence located on the end point side, and the weight v between the sentences are described as a pair. If it becomes, it will become a text network of FIG.

Here, the inter-sentence weight “5” is assigned to the directed line segment designated by the teaching information 14B from the start point sentence D163 to the end point sentence D357.
Further, as new directed line segments other than those specified by the teaching information 14B, directed line segments from the start point sentence D163 to the end point candidate sentences D380 and D421 are generated, and the inter-sentence weight is “12”. "," 7 ". In addition, directed line segments from the start point candidate sentence D196 to the end point sentence D357 and the end point candidate sentences D380 and D421 are generated, and the inter-sentence weights are “10”, “8”, and “6”, respectively. .

In this way, when generating a new directed line segment in the sentence network, the directed line segment specified in the teaching information is decomposed into a directed line segment between important words included in the start point sentence and the end point sentence. Generating important word network information consisting of directed line segments having weights between important words, and starting point candidate sentence and end point candidate sentences used for generating new directed line segments based on the important word network information Therefore, not only the presence / absence of each sentence and its order, but also the weight of the relation can be expressed by the sentence network.
Thereby, when searching related knowledge for an arbitrary task, it is possible to easily search for knowledge that has a large connection to the task based on the weight between sentences, and it is possible to efficiently search for desired knowledge from a lot of knowledge.

  Further, according to the equation (2) for calculating the important word weight w, when the teaching is repeated for the same important inter-word directed line segment, the important inter-word weight w corresponding to the directed line segment. Since the relationship between the start point important word and the end point important word becomes stronger, it is possible to accumulate valid teachings in the key word network. In addition, the coefficient γ may be expressed as a time function so that the weights between important words attenuate with the passage of time in the important word network. Relevance can be reduced and new relevance can be used effectively.

Next, with reference to FIG. 12, a knowledge search operation in the related knowledge search device according to the third exemplary embodiment of the present invention will be described. FIG. 12 is a flowchart illustrating a knowledge search process in the related knowledge search apparatus according to the third embodiment.
In the present embodiment, a case will be described in which a sentence related to a desired sentence is searched using the sentence network information 14D obtained by the sentence network generation operation according to the second embodiment.

The arithmetic processing unit 15 starts the knowledge search process of FIG. 12 by the knowledge search unit 15F in response to a predetermined related knowledge search instruction operation from the operation input unit 12.
First, the knowledge search means 15F acquires the specification of the search target sentence from the operation input unit 12 as a task (step 130), and from the sentence network information 14D of the storage unit 14, for each directed line segment that uses the search target sentence as the starting point sentence. Then, the end point sentence and the inter-sentence weight are acquired (step 131). Then, the screen display unit 11 displays only the obtained end point sentences in descending order of the inter-sentence weight or those having the inter-sentence weight larger than a predetermined threshold (step 132).

  Subsequently, the knowledge search unit 15F acquires the start point sentence and the inter-sentence weight for each directed line segment having the search target sentence as the end point sentence from the sentence network information 14D in the storage unit 14 (step 133). Then, the screen display unit 11 displays only the obtained starting point sentences in descending order of the inter-sentence weight or those having the inter-sentence weight greater than a predetermined threshold (step 134).

  Therefore, when knowledge related to the sentence D163 is searched using the sentence network information shown in FIGS. 10 and 11, the sentence D163 specified by the teaching information 14B as shown in the search result screen display example of FIG. In addition to the directed line segment from the text D357 to the text D357, new directed line segments from the text D163 to the texts D380 and D421 are obtained. At this time, it is displayed from the top in descending order of the weight between each sentence, and it can be easily understood that there is a sentence more relevant than the directed line segment designated by the teaching information 14B.

  When the knowledge related to the sentence D163 is searched using the sentence network information shown in FIGS. 10 and 11, the sentence D163 specified by the teaching information 14B as shown in the search result screen display example of FIG. In addition to the directed line segment from the sentence D357 to the sentence D357, a new directed line segment having the sentence D357 as the end point is obtained. At this time, it is displayed from the top in the descending order of the weight between the sentences, and it can be easily understood that there is a sentence more important than the directed line segment specified by the teaching information 14B.

In the above, depending on a predetermined sentence network generation instruction operation from the operation input unit 12 or completion of reception of new teaching information 14B in the teaching information receiving unit 15A, a sentence network generation process or an important word network generation process. Although the case where the sentence network generation processing is performed has been described as an example, the processing timing is not limited to the above.
For example, the teaching information receiving unit 15A sequentially receives teachings and sequentially stores them in the teaching information 14B of the storage unit 14, and when there is an associated knowledge search instruction operation from the operation input unit 12, the storage unit 14 Based on the teaching information 14B, the sentence network generation process, or the keyword network generation process and the sentence network generation process may be executed. As a result, when the related knowledge search process is not performed, the sentence network generation process and the keyword network generation process are not performed, and unnecessary processes can be omitted.

  In the above description, the related knowledge search device having the knowledge search means as the function for searching the related knowledge has been described as an example. However, the related knowledge search network that generates only the sentence network information or the keyword network information used for the related knowledge search. An information generation device can also be configured.

In this case, as the related knowledge search network information generation device corresponding to the first embodiment described above, the arithmetic processing unit 15 in the configuration of FIG. 1 includes at least the sentence network generation unit 15B and the sentence weight calculation. Means 15C may be provided, and the storage unit 14 may be provided with a sentence set 14A, teaching information 14B, and an important word list 14C as processing information.
In addition to the above configuration, the related knowledge search network information generating apparatus corresponding to the second embodiment described above includes an important word network generating unit 15D and an important word weight calculating unit 15E in the arithmetic processing unit 15. It should be provided.

It is a block diagram which shows the structure of the related knowledge search device concerning one embodiment of this invention. It is a structural example of a text set. It is a structural example of an important word list. It is a flowchart which shows the text network production | generation process concerning the 1st Embodiment of this invention. It is explanatory drawing which shows the example of a text network production | generation operation | movement concerning the 1st Embodiment of this invention. It is a structural example of the text network information obtained by the text network generation process of FIG. It is a flowchart which shows the important word network production | generation process concerning the 2nd Embodiment of this invention. It is a structural example of the important word network information obtained by the important word network generation processing of FIG. It is a flowchart which shows the text network production | generation process concerning the 2nd Embodiment of this invention. 10 is a configuration example of text network information obtained by the text network generation process of FIG. 9. It is a graph which shows the text network of FIG. It is a flowchart which shows the knowledge search process concerning the 3rd Embodiment of this invention. 13 is a search result screen display example obtained by the knowledge search process of FIG. 12. 13 is another search result screen display example obtained by the knowledge search process of FIG. 12.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Related knowledge search device, 11 ... Screen display part, 12 ... Operation input part, 13 ... Information input / output part, 14 ... Memory | storage part, 14A ... Text set, 14B ... Teaching information, 14C ... Important word list, 14D ... Text Network information, 14E ... Keyword network information, 14F ... Program, 15 ... Calculation processing unit, 15A ... Teaching information receiving means, 15B ... Sentence network generating means, 15C ... Intertext weight calculating means, 15D ... Keyword network generating means, 15E: Important word weight calculation means, 15F: Knowledge search means.

Claims (10)

A storage unit for storing a sentence set composed of a plurality of sentences in which various kinds of knowledge are described in character information using a predetermined important word, and a sentence network information indicating a relation between the sentences belonging to the sentence set, and A related knowledge search device having an arithmetic processing unit for searching a desired sentence related to a sentence including knowledge to be searched from a sentence set of the storage unit using sentence network information,
The arithmetic processing unit receives inter-sentence relationship information indicating relevance and order between two sentences, and inputs the start-point sentence located on the start point side of the inter-sentence relation information and the end point side of the inter-sentence relation information. A sentence network generating means for generating the sentence network information based on an important word included in the located end point sentence;
The sentence network generation means selects a sentence including all the same important words as the start sentence specified by the input sentence relation information as a start point candidate sentence from the sentence set of the storage unit, and the sentence relation A sentence including all the same important words as the end sentence specified by information is selected as an end point candidate sentence from the sentence set of the storage unit, and the end point sentence and the end point are respectively set as the start point sentence and the start point candidate sentence. Generate new inter-sentence relationship information with each candidate sentence as an end point, and from the pair of the information indicating the start point sentence and the information indicating the end point sentence of the new inter-sentence relation information, A related knowledge retrieval device that generates sentence network information indicating order. The related knowledge search device according to claim 1,
The arithmetic processing unit, as an arbitrary sentence relationship information, each of the important words included in the start sentence of the inter-sentence relation information as the start point, and each of the important words included in the end point sentence of the inter-sentence relation information as the end point Further comprising a sentence weight calculation means for calculating the sentence weight corresponding to the sentence relation information based on the sum of the important word weights corresponding to the important word relation information. ,
The related knowledge search device, wherein the sentence network generation unit generates a sentence network having an inter-sentence weight obtained by the inter-sentence weight calculation unit for each sentence. The related knowledge search device according to claim 2,
The arithmetic processing unit is configured to input each of the important words included in the end sentence of the inter-sentence relation information from each of the important words included in the start sentence of the inter-sentence relation information. It further comprises an important word weight calculation means for decomposing each important word relation information as an end point and updating the important word weight corresponding to the important word relation information for each occurrence of the important word relation information. A related knowledge retrieval device as a feature. The related knowledge search device according to claim 1,
The arithmetic processing unit retrieves a sentence having a sentence including desired knowledge as a start sentence or an end sentence based on the sentence network information from a sentence set in the storage unit, and outputs the sentence as a sentence related to the knowledge The related knowledge search device, further comprising: The related knowledge search device according to claim 2,
The arithmetic processing unit searches the sentence set of the storage unit for a sentence having a sentence including desired knowledge as a start point sentence or an end point sentence based on the sentence network information, and obtains the obtained sentence in the order of the weight between the sentences. A related knowledge search apparatus, further comprising knowledge search means for outputting as a sentence related to knowledge. A storage unit that stores a set of sentences composed of a plurality of sentences in which various kinds of knowledge are described in character information using a predetermined important word, and a desired sentence related to a sentence including the search target knowledge is searched from the set of sentences in the storage unit. A sentence network information generating device having an arithmetic processing unit for generating sentence network information used to
The arithmetic processing unit receives inter-sentence relationship information indicating relevance and order between two sentences, and inputs the start-point sentence located on the start point side of the inter-sentence relation information and the end point side of the inter-sentence relation information. A sentence network generating means for generating the sentence network information based on an important word included in the located end point sentence;
The sentence network generation means selects a sentence including all the same important words as the start sentence specified by the input sentence relation information as a start point candidate sentence from the sentence set of the storage unit, and the sentence relation A sentence including all the same important words as the end sentence specified by information is selected as an end point candidate sentence from the sentence set of the storage unit, and the end point sentence and the end point are respectively set as the start point sentence and the start point candidate sentence. Generate new inter-sentence relationship information with each candidate sentence as an end point, and from the pair of the information indicating the start point sentence and the information indicating the end point sentence of the new inter-sentence relation information, A sentence network information generating apparatus for generating sentence network information indicating order. A storage unit that stores a set of sentences composed of a plurality of sentences in which various kinds of knowledge are described in character information using a predetermined important word, and a desired sentence related to a sentence including the search target knowledge is searched from the set of sentences in the storage unit. A sentence network generation method for generating sentence network information based on inputted predetermined sentence relationship information, which is used in a sentence network information generation apparatus having an arithmetic processing unit for generating sentence network information to be used,
In the arithmetic processing unit, the inter-sentence relationship information indicating the relationship and order between the two sentences is input, and the start point sentence located on the start point side of the inter-sentence relationship information and the end point side of the inter-sentence relationship information A sentence network generating step for generating the sentence network information based on an important word included in the end point sentence located;
In this sentence network generation step, a sentence including all the same important words as the starting point sentence specified by the inputted inter-sentence relation information is selected as a starting point candidate sentence from the sentence set of the storage unit, and the inter-sentence relation Selecting a sentence including all the same important words as the end sentence specified by information as an end point candidate sentence from the sentence set of the storage unit;
Information indicating the starting point sentence of the new inter-sentence relation information, and generating new inter-sentence relation information starting from the starting point sentence and the starting point candidate sentence and starting from the end point sentence and the end point candidate sentence, respectively. And a step of generating sentence network information indicating the relevance and order between the sentences from a pair of information indicating the end sentence and the sentence network generating method. The sentence network generation method according to claim 7,
In the arithmetic processing unit, any inter-sentence relationship information, each of the important words included in the start point sentence of the inter-sentence relation information as the start point, and each of the important words included in the end point sentence of the inter-sentence relation information as the end point Further comprising a sentence weight calculation step for calculating a sentence weight corresponding to the sentence relation information based on a sum of the weights of the important words corresponding to the important word relation information. ,
The sentence network information generation step generates a sentence network having the sentence weight obtained in the sentence weight calculation step for each sentence. The sentence network generation method according to claim 7,
In the arithmetic processing unit, each of the important words included in the end sentence of the inter-sentence relation information is determined by using the input inter-sentence relation information as the start point of each of the important words included in the start sentence of the inter-sentence relation information. It further comprises a significant word weight calculation step for decomposing each important word relation information as the end point and updating the important word weight corresponding to the important word relation information for each occurrence of the important word relation information. Character network generation method characterized. A storage unit that stores a set of sentences composed of a plurality of sentences in which various kinds of knowledge are described in character information using a predetermined important word, and a desired sentence related to a sentence including the search target knowledge is searched from the set of sentences in the storage unit. A computer of a text network information generation device having a calculation processing unit for generating text network information used to
The sentence relation information indicating the relationship and order between the two sentences is input, and a sentence including the same important word as the start sentence located on the start point side of the sentence relation information is started from the sentence set in the storage unit. Selecting as a candidate sentence and selecting a sentence including the same important word as the end point sentence located on the end point side of the inter-text relation information from the sentence set of the storage unit as an end point candidate sentence;
Information indicating the starting point sentence of the new inter-sentence relation information, and generating new inter-sentence relation information starting from the starting point sentence and the starting point candidate sentence and starting from the end point sentence and the end point candidate sentence, respectively. And a step of generating sentence network information indicating relevance and order between the respective sentences from a pair of information indicating the end sentence and the sentence.

Images