JP5034580B2 - Causal reasoning apparatus, control program and control method thereof - Google Patents

Description

  The present invention relates to a causal reasoning apparatus that supports a search for a cause associated with a target phenomenon, a control program thereof, and a control method thereof.

  2. Description of the Related Art Conventionally, a search method is known in which target information is selected from a database in which various types of information are stored based on surface layer similarities such as character strings, words, and images.

  In recent years, attempts have been made to use search in order to search for causes that lead to a target phenomenon. Specifically, for the purpose of investigating the cause of failure, attempts have been made to search for cases from a database of past failure cases.

In addition, for example, Patent Document 1 discloses a failure diagnosis apparatus that searches for a cause of a failure associated with a failure phenomenon such as a vehicle or an aircraft. In other words, when past cases including a predetermined defect phenomenon are stored, and a character string representing the defect phenomenon is input as a search character string, the defect represented by the character string most similar to the input character string There is disclosed a failure diagnosis apparatus that selects a case including a phenomenon and infers a cause according to a predetermined rule from a failure phenomenon included in the selected case.
JP 7-271590 A (published on October 20, 1995) “Extraction of technical terms based on appearance frequency and connection frequency”, natural language processing, vol.0, No.0, 2001 “Chunk Identification Using Support Vector Machine”, Natural Language Processing, vol.2, No.3, 1995

  For the purpose of investigating the cause of failure, assuming a search using a conventional search technique from a database of past failure cases, search using a character string representing the target phenomenon as a search character string. The past cases where only the phenomenon is similar are selected by the case search. However, the cause obtained from the past case where the phenomenon is similar is not necessarily the cause of the target phenomenon. An example of a case where a search is performed from a database in which past failure cases are collected using the above-described conventional search method will be described with reference to FIG. ”,“ Leakage ”, and“ Ignition ”are common, and the phenomenon is similar, but the causes of“ Ignition ”are“ Pipe deterioration ”and“ Valve clogging due to increased oil viscosity ”, respectively. Is different.

  As described above, in the conventional search method, there is a high possibility that a case with a low similarity will be selected for the causal relationship between the phenomenon and the cause, which is important for investigating the cause of the failure. There is a high possibility that the cause cannot be selected. That is, there is a possibility that the search accuracy may be lowered in the case search for finding the cause that leads to the target phenomenon.

  Further, in the failure diagnosis apparatus disclosed in Patent Document 1, when a character string representing an observable defect phenomenon is used as a search character string, the past includes a defect phenomenon having a surface similarity such as a character string similarity. The case is selected and the cause is inferred based on the selected case. Therefore, in the fault diagnosis apparatus disclosed in Patent Document 1, there is a possibility of inferring the cause of the causal relationship between the phenomenon and the cause, which is important for investigating the cause of the failure, based on a case of low similarity. Get higher. That is, there is a high possibility that an erroneous cause is inferred based on an incorrect case. Therefore, even in the failure diagnosis apparatus disclosed in Patent Document 1, there is a possibility that the search accuracy may be lowered in the case search for finding the cause that leads to the target phenomenon.

  The present invention has been made in view of the above-described conventional problems, and the purpose thereof is a causal reasoning apparatus that can further improve search accuracy in case search for obtaining a cause that leads to a target phenomenon. And providing a control program and a control method thereof.

  In order to solve the above problem, the causal reasoning apparatus of the present invention, by analyzing sentence data acquisition means for acquiring sentence data described about the phenomenon of interest, and sentence structure in the received sentence, Case data is obtained from a first causal relationship extracting means for extracting a causal relationship between words in a sentence included in the sentence, and a database storing a plurality of case data that are sentences written in advance about the phenomenon. Case data acquisition means, second causal relation extraction means for extracting the causal relationship between words in the sentence included in the sentence in the case by analyzing the sentence structure in the sentence of the acquired case, and second causal relation Each of a plurality of causal relationships extracted from a plurality of cases by the extraction means is connected in accordance with the overlap between words constituting the causal relationship and expressed in a network form. Based on the causal network creation means for creating the causal network data, the causal relation data extracted by the first causal relation extraction means, and the causal network data, on which path on the causal network the causal relation is located. Inference means for inferring whether they are similar and generating a causal path based on the inferred path is provided.

  In addition, in order to solve the above-described problem, the control method of the present invention includes a sentence data acquisition step of acquiring sentence data describing a target phenomenon by a sentence data acquisition unit, and a first causal relationship extraction unit. By analyzing the sentence structure in the sentence received in the sentence data obtaining step, the first causal relation extracting step for extracting the causal relation between words in the sentence included in the sentence and the case data obtaining means A case data acquisition step for acquiring case data from a database storing a plurality of case data, which are sentences written in advance, and a case sentence acquired in the case data acquisition step by the second causal relationship extraction means Analyzing the sentence structure in the sentence, the causal relationship between words in the sentence included in the sentence in the case is extracted. The causal relationship extracting step and the causal network creating means connect the plurality of causal relationships extracted from the plurality of cases by the second causal relationship extracting step according to the overlap between the words constituting the causal relationship, The causal network creation step for creating the causal network data, and the causal network data extracted in the first causal relationship extraction step by the inference means and the causal network data. It is characterized by including an inference step for inferring which path on the causal network is similar to the relation and generating a causal path based on the inferred path.

  According to said invention, the causal relationship between the phrases in the sentence contained in the said sentence is extracted by the 1st causal relationship extraction means from the text which received the input by the text data acquisition means. In addition, the second causal relationship extracting unit extracts the causal relationship between the phrases in the sentence included in the case from the plurality of cases acquired by the case data acquiring unit, and each of the extracted plurality of causal relationships is converted into the causal relationship. The data of the causal network is created by connecting according to the overlap between the words constituting the word. Then, a causal path corresponding to a result of inferring which path on the causal network is similar to the causal relation extracted by the first causal relation extracting means is generated by the inference means.

  A path on the causal network that is more similar to the causal relationship extracted by the first causal relation extracting unit is inferred by inferring to which path on the causal network the causal relationship extracted by the first causal relationship extracting unit is similar. It becomes possible to ask. Then, based on the path on the causal network similar to the causal relationship extracted by the first causal relation extracting unit, the causal similar to the path on the causal network similar to the causal relationship extracted by the first causal relationship extracting unit. If a path is generated, the causal network consists of causal relationships obtained from a plurality of cases that are sentences written in advance about the phenomenon. A case can be selected from the plurality of cases.

  Accordingly, it is possible to more appropriately select cases including causes that are linked to the target phenomenon by searching. As a result, there is an effect that it is possible to further improve the search accuracy in the case search for finding the cause associated with the target phenomenon.

  By the way, the causal reasoning apparatus may be realized by causing a computer to execute a program. Specifically, the control program according to the present invention is a program that causes a computer to operate as the causal reasoning apparatus.

  When these control programs are executed by a computer, the computer operates as the causal inference apparatus. Therefore, similarly to the above-described causal reasoning apparatus, it is possible to further improve the search accuracy in the case search for obtaining the cause associated with the target phenomenon.

  In the causal inference apparatus of the present invention, the causal relation data collecting means for collecting the causal relation data extracted by the second causal relation extracting means for each case, and the causal relation data collected for each case. The causal relationship data storage unit that stores the causal relationship data group that is distinguished for each case and the causal relationship data group that is similar to the causal path generated by the inference means are selected from the causal relationship data storage unit and selected. It is preferable that the information processing apparatus further includes similarity search means for reading a case corresponding to the causal relationship data group from the database.

  As a result, a causal data group similar to the causal path generated by the inference means is selected from the causal relation data storage unit by the similarity search means. In the causal relation data storage unit, the causal relation data group collected by the causal relation data collecting means is stored separately for each case, so that the case corresponding to the selected causal relation data group is read from the database. Is possible. In addition, if a causal path similar to the path on the causal network similar to the causal relationship extracted by the first causal relation extracting unit is used as the causal path, a case similar to the sentence that has received the input and the causal relation is selected. It is possible to select from cases.

  Accordingly, it is possible to more appropriately select cases including causes that are linked to the target phenomenon by searching. As a result, there is an effect that it is possible to further improve the search accuracy in the case search for finding the cause associated with the target phenomenon.

  In the causal inference apparatus according to the present invention, it is preferable that the causal path is a path on the causal network corresponding to a location where the causal relationship extracted by the first causal relationship extracting unit is similar to the causal network.

As a result, the causal reasoning apparatus can use the path on the causal network corresponding to the location where the causal relation extracted by the first causal relation extracting unit and the causal network most closely match as a causal path. In the causal inference apparatus, the causal path is a path obtained by connecting the causal relation to a path on the causal network corresponding to a place where the causal relation extracted by the first causal relation extracting unit is similar to the causal network. Preferably there is.

  As a result, the causal inference apparatus uses a path obtained by connecting the causal relation to the path on the causal network corresponding to the location where the causal relation extracted by the first causal relation extracting unit most closely matches the causal network. Can be used

  According to the present invention, the causal relationship extracted by the first causal relationship extracting unit is the most inferred by the reasoning unit in which the causal relationship extracted by the first causal relationship extracting unit is similar to the path on the causal network. It is possible to obtain a similar path on the causal network. And similar to the path on the causal network similar to the causal relationship extracted by the first causal relation extracting means according to the path on the causal network more similar to the causal relation extracted by the first causal relation extracting means. If the cause-and-effect path is generated, the cause-and-effect network is composed of a plurality of cases that are sentences written in advance about the phenomenon. It is possible to select from cases. Accordingly, it is possible to more appropriately select cases including causes that are linked to the target phenomenon by searching. As a result, there is an effect that it is possible to further improve the search accuracy in the case search for finding the cause associated with the target phenomenon.

  An embodiment of the present invention will be described with reference to FIGS. 1 to 13 as follows.

  First, the outline of the configuration of the causal inference apparatus 1 will be described with reference to FIG. FIG. 1 is a functional block diagram showing the configuration of the causal reasoning apparatus 1 according to the present embodiment. As shown in FIG. 1, the causal inference apparatus 1 includes an input unit (sentence data acquisition means) 11, a first important word extraction unit 12, a first generalization processing unit 13, a sentence structure analysis unit (first causal relationship extraction). Means) 14, case data acquisition unit (case data acquisition unit) 15, second important word extraction unit 16, second generalization processing unit 17, causal relationship extraction unit (second causal relationship extraction unit) 18, causal network creation unit (Causal network creation means) 19, causal network data storage section 20, causal relation data collection section (causal relation data collection means) 21, causal relation data storage section (causal relation data storage section) 22, and inference section 23. Yes. The inference unit 23 includes a matching unit (inference unit) 31 and a similar search unit (similar search unit) 32.

  First, the input unit 11 receives an input of a sentence (search character string) including one or more sentences described about a target phenomenon, and includes various keys. The input unit 11 sends text data of a sentence that has been accepted to the first important word extraction unit 12. Note that the input unit 11 may receive text (search character string) data sent from a device external to the causal reasoning device 1 by wire or wirelessly.

  The first important word extraction unit 12 is stored in the domain dictionary from the text data sent from the input unit 11 based on the domain dictionary data stored in the external device of the causal reasoning device 1. A sentence including a certain word (important word) is selected and extracted. The first important word extraction unit 12 may be configured to calculate the importance of a word based on the syntax analysis result and extract a sentence including a word (important word) having a high importance. That is, the 1st important word extraction part 12 may be the structure which extracts the sentence containing an important word by the well-known automatic technical vocabulary extraction method used by "word selection Web" etc., for example. Then, the first important word extraction unit 12 sends text data of a sentence including the extracted important words to the first generalization processing unit 13. The domain dictionary referred to here is a collection of terms in a specific field. In the present embodiment, a sentence including an important word is selected, but a phrase including an important word may be selected and a subsequent process may be performed.

  The first generalization processing unit 13 generalizes terms in each sentence including important words based on the data of synonym dictionaries provided in a device external to the causal reasoning device 1. Further, the first generalization processing unit 13 sends text data of each sentence including the generalized word (term text data of the generalized sentence) to the sentence structure analysis unit 14. . The synonym dictionary is a grouping of words in synonym relations and synonym relations. The term generalization referred to here is a process of converting words in synonym relations and synonym relations into unified terms. For example, a thesaurus may be used as the thesaurus.

  The sentence structure analysis unit 14 uses sentence structure knowledge data provided in an external device of the causal reasoning apparatus 1 to generalize sentence structures such as case analysis and dependency analysis of each sentence that has been generalized. Analyze. The sentence structure analysis unit 14 generates text data to which information on the sentence structure of each sentence is given, and sends it to the inference unit 23 as a result of the analysis of the sentence structure. The sentence structure knowledge mentioned here includes syntax structure, grammar knowledge, dependency knowledge, case structure knowledge, and the like. In the present embodiment, case analysis and dependency analysis represent processing for determining the relationship between terms in each sentence. The relationship between the terms in each sentence determined by the case analysis and dependency analysis includes, for example, a connection / connection relationship. For example, as a result of sentence structure analysis, if there is a tangent relationship between terms in the sentence, the relationship between cause and effect is established between the terms. It is possible to obtain data on the relationship (causal relationship) between the event corresponding to the cause and the event corresponding to the result. In this embodiment, the process of determining the relationship between terms in each sentence is performed. However, the process of determining the relationship between clauses in each sentence and performing subsequent processes. There may be.

  Each of the first important word extraction unit 12, the first generalization processing unit 13, and the sentence structure analysis unit 14 receives data such as a domain dictionary, a thesaurus, and a sentence structure knowledge, a domain dictionary, a thesaurus, and a sentence structure knowledge. It is used by accessing a device such as an external storage device that stores data such as a server connected via a network.

  The case data acquisition unit 15 obtains past case data (case data) from a case database (a database storing a plurality of case data, which are sentences written in advance regarding phenomena), which stores past case data. To get. The case data acquisition unit 15 sends the acquired case data to the second important word extraction unit 16. Subsequently, the second important word extraction unit 16 converts the case data sent from the case data acquisition unit 15 into the domain dictionary based on the domain dictionary data stored in the external device of the causal reasoning device 1. A sentence including the stored word (important word) is selected and extracted. Then, the second important word extraction unit 16 sends text data of a sentence including the extracted important word to the second generalization processing unit 17.

  The second generalization processing unit 17 generalizes terms in each sentence including important words based on the data of synonym dictionaries provided in a device external to the causal reasoning device 1. Further, the second generalization processing unit 17 sends the text data of each sentence including the important word that has been generalized the term to the causal relationship extraction unit 18.

  The causal relation extraction unit 18 uses sentence structure knowledge data provided in a device external to the causal reasoning apparatus 1 to analyze sentence structures such as case analysis and dependency analysis of each sentence that has been generalized. Analysis is performed, and the relationship between terms in each sentence determined by analyzing the sentence structure is extracted as a relationship (causal relationship) between an event that is a cause and an event that is a result. In the present embodiment, the processing for determining the relationship between terms in each sentence is performed. However, the sentence structure analysis unit 14 determines the relationship between clauses in each sentence, and If the configuration is such that the causal relationship extraction unit 18 determines the relationship between clauses in each sentence, the subsequent processing is performed.

  Here, the extraction of the causal relationship will be described with reference to FIGS. 2 (a) and 2 (b). FIG. 2A is a diagram illustrating an example of a result of extracting sentences including multiple terms from each of a plurality of cases. In FIG. 2 (a), a part of the result of extracting a sentence including multiple terms is extracted for each number for identifying a case. FIG. 2B is a diagram illustrating an example of the causal relationship extracted by the causal relationship extracting unit 18. In FIG. 2B, the causal relationship is expressed by drawing an arrow from the event that is the cause to the event that is the result.

  For example, when a specific example is shown by using the data of case number 79 in FIG. 2A, first, a sentence including a multiple term extracted from case number 79 “The ductility of the steel sheet has been lost due to a rapid temperature drop”. With the generalization of the term, for example, “rapid temperature drop” is converted to “temperature drop”, and “steel sheet ductility is lost” is converted to “material deterioration”. And by analyzing the sentence structure of the sentence that was converted to “material deterioration due to temperature drop” due to the generalization of terms, the causal relationship that “temperature deterioration” resulted in “material deterioration” Extract. The causal relationship extraction unit 18 performs the above causal relationship extraction processing for a plurality of cases.

  Each of the second important word extraction unit 16, the second generalization processing unit 17, and the causal relationship extraction unit 18 receives data such as a domain dictionary, a thesaurus, and a sentence structure knowledge, a domain dictionary, a thesaurus, and a sentence structure knowledge. It is used by accessing a device such as an external storage device that stores data such as a server connected via a network.

  The causal network creation unit 19 uses a plurality of causal relationship data extracted by the causal relationship extraction unit 18 so that an event corresponding to the result of one causal relationship matches an event corresponding to the cause of the other causal relationship. A causal network is created by connecting each other in the form of sharing matching events. In addition, the causal network creation unit 19 is configured by combining a causal relationship extracted by the causal relationship extraction unit 18 with an event that causes a certain cause among all the extracted causal relationships. By determining the probability, the conditional probability (causal strength) that the event that occurs when the event that causes it occurs will be obtained and assigned to each causal relationship. In the present embodiment, the causal strength (score) is represented by a value of 0.0 to 1.0.

  In the present embodiment, the causal network creation unit 19 obtains and assigns the causal strength of each causal relationship. However, the present invention is not limited to this. For example, by displaying the causal relationship data extracted by the causal relationship extracting unit 18 on a display unit (not shown), a score setting unit for each causal relationship from the user is received by a score setting unit (not shown) and set by the user. The structure which assign | provided score to each corresponding causal relationship may be sufficient.

  The causal network data storage unit 20 stores the causal network data created by the causal network creation unit 19. For example, if a causal network is created based on a plurality of past case data for a specific field, it is possible to cover the causal relationship of the specific field.

  Here, the causal network will be described with reference to FIGS. 3 (a) and 3 (b). FIG. 3A is a diagram illustrating an example of a causal network in the present embodiment. FIG. 3B is a diagram illustrating an example of association between the causal relationship and the causal strength (causal strength).

  As shown in FIG. 3A, the causal network shows a relationship between a causal event and a resulting event as a directed graph with a score. In the causal network in the present embodiment, a plurality of nodes exist in the middle of a path (causal path) from a certain event to another certain event. For example, in FIG. 3A, there are nodes such as “material degradation”, “crack”, and “leakage” in the middle of the path from the event node “Aging” to the event node “Fire”. Yes. Each node indicates a specific event, and an event corresponding to the cause and an event corresponding to the result among the causal relationships extracted by the causal relationship extracting unit 18 are assigned. For example, in the causal relationship between “material degradation” and “crack”, “material degradation” is the cause and “crack” is the event, but in the causal relationship between “crack” and “leakage”, The structure is such that the relationship between the upstream and downstream of the causal path is always the relationship between the cause and the result, such that “break” is the cause of the event and “leakage” is the event of the result. In the present embodiment, as shown in the table of FIG. 3B, a score indicating a conditional probability (causal strength) that a resultant event occurs when a causal event occurs is a cause. It is assumed that it is given to an arc connecting the node corresponding to the event to be and the node corresponding to the event to be the result. When a specific example is shown by taking the table of FIG. 3B as an example, if the event (cause) that is the cause is “Aged”, the causal strength of “material deterioration” of the event (result) that is the result Becomes 1.0. Further, when the cause event (cause) is “insufficient maintenance check”, the causal strength of the resulting event (result) with “age” is 0.7. In the present embodiment, the causal strength is represented by a value of 0.0 to 1.0, but is not necessarily limited thereto, and may be represented by, for example, a percentage (%).

  Subsequently, the causal relationship data collection unit 21 collects the causal relationship data extracted by the causal relationship extraction unit 18 for each case derived from the causal relationship data. For each data group), an ID for identifying corresponding case data is assigned and stored in the causal relationship data storage unit 22. In addition, the causal relationship data storage unit 22 stores the causal relationship data group collected by the causal relationship data collection unit 21, and the data stored in the causal relationship data storage unit 22 is a causal relationship data group. Each ID is assigned to identify the case data corresponding to the data.

  The inference unit 23 matches the result of the sentence structure analysis (causal relation data) sent from the sentence structure analysis unit 14 with the causal network data stored in the causal network data storage unit 20, Similarity search is performed by using causal paths on the causal network with overlapping overlap. Matching here refers to comparing causal data with causal network data and selecting (generating) a causal path on the causal network that is similar to the causal data (most coincident). Represents what to do. The similarity search represents obtaining a case including a causal relationship similar to the causal path by searching a causal relationship data group including a causal relationship similar to the causal path selected by the matching unit 31. The matching may be configured such that the causal relation data and the causal network data are compared, and the causal path on the causal network that most closely matches the causal relation data is selected (generated). .

  Further, the matching unit 31 included in the inference unit 23 is based on the causal network data stored in the causal network data storage unit 20 and the causal network data sent from the sentence structure analysis unit 14. Matching is performed. Then, the similarity search unit 32 included in the inference unit 23 uses the causal relationship data group stored in the causal relationship data storage unit 22 based on the causal path data sent from the matching unit 31 to the causal relationship data group. It searches for a causal relationship data group including a causal relationship similar to a path. Further, the similarity search unit 32 reads cases corresponding to the causal relationship data group obtained by the search from the case database, and outputs the read cases as search results to a display unit (not shown). Note that a display unit (not shown) may be provided in the causal inference apparatus 1 or may be provided outside the causal inference apparatus 1. The similarity search unit 32 reads a case corresponding to a causal relationship data group including a causal relationship most similar to the causal path from the causal relationship data group stored in the causal relationship data storage unit 22 from the case database. It may be a configuration in which cases corresponding to a causal relationship data group including a causal relationship whose similarity with the causal path is equal to or higher than a predetermined height are read from the case database in descending order of the similarity. Good. Note that the similarity here refers to a score based on a ratio that matches the causal path. As a specific example, when A, B, C, D, E, and F are nodes, the similarity between the causal path ABCDE and the causal path ABFDE is that the matching paths are AB and DE. There are two paths that do not match: BC and CD (or BF and FD). Note that the rate of matching with the causal path is obtained by dividing the number of matching paths by the total number of paths, so the rate of matching with the causal path, that is, the similarity is 2 ÷ 4 = 0.5. The predetermined height is arbitrarily set.

  In the present embodiment, a configuration including an automatic dictionary creation unit that creates a dictionary by statistical processing based on the important words extracted by the second important word extraction unit 16 may be used. Based on the data of the dictionary created by the statistical processing created by the section, the first important word extracting unit 12 selects the words (important words) stored in the statistically processed dictionary from the text data sent from the input unit 11. The structure which selects the sentence to include and extracts may be sufficient.

  Note that the process of automatically generating dictionary data by statistical processing based on important words extracted from case data is realized by using a known automatic term extraction method described in Non-Patent Document 1, for example. The structure which makes it possible may be sufficient.

  In the present embodiment, the data of the sentence structure knowledge of the data of each sentence including the important word based on the data of each sentence including the important word subjected to the term generalization in the second generalization processing unit 17 The first generalization processing unit 13 may use an important word based on the sentence structure knowledge data created by the sentence structure knowledge automatic creation unit. The structure which generalizes the term in each sentence containing can be sufficient. The process for automatically generating sentence structure knowledge may be realized by using a known method described in Non-Patent Document 2, for example.

  In the present embodiment, the causal relationship data collecting unit 21 collects the causal relationship data for each case, but the present invention is not necessarily limited thereto. For example, the causal relationship data collection unit 21 creates a causal network for each case to which a causal network for each case or no score is assigned from the causal relationship data collected collectively for each case, and the causal relationship data storage unit 22 may be configured to store data. In this case, the matching unit 31 performs matching between the causal network data stored in the causal network data storage unit 20 and the causal network data for each case stored in the causal relationship data storage unit 22, and the most overlapping A similar search may be performed using a causal path on the matched causal network (causal network stored in the causal network data storage unit 20).

  Further, in the present embodiment, the configuration in which the similarity search unit 32 performs the similarity search using the causal path data obtained by the matching unit 31 has been described, but the configuration is not necessarily limited thereto. For example, the causal path data obtained by the matching unit 31 may be sent to a display unit (not shown), and the causal path may be displayed on the display unit.

  Next, the operation flow in the causal inference apparatus 1 will be described with reference to FIGS. FIG. 4 is a flowchart showing an operation flow in the causal reasoning apparatus 1. FIG. 5 is a diagram illustrating an example of text input to the input unit 11. FIG. 6 is a diagram illustrating an example of selection of a sentence including an important word by the first important word extraction unit 12. FIG. 7 is a diagram illustrating an example of generalization of terms by the first generalization processing unit 13. FIG. 8 is a diagram illustrating an example of a sentence structure analysis result by the sentence structure analysis unit 14.

  First, in step S1, the input unit 11 accepts input of text (search character string) or text data, and proceeds to step S2. As a specific example, in the process of step S1, the input unit 11 receives an input of a sentence as a source sentence of the subsequent process as shown in FIG. Then, in step S2, the first important word extraction unit 12 uses the domain dictionary data, and the sentence including words (important words) stored in the domain dictionary from the text data sent from the input unit 11 Select and extract. As a specific example, in the process of step S2, as indicated by the underlined portion in FIG. 6, only the sentence including the important word is selected by the first important word extracting unit 12, and the text data of the sentence including the important word is obtained. It is sent to the first generalization processing unit 13.

  Subsequently, in step S3, the first generalization processing unit 13 generalizes terms in each sentence including important words based on the data of the thesaurus, and proceeds to step S4. As a specific example, in the process of step S3, as shown in the shaded part of FIG. 7, based on the data of the thesaurus, if there are terms that have a synonym relationship, a synonym relationship, etc., The first generalization processing unit 13 performs conversion into the unified term, and the generalized text data of the sentence is sent to the sentence structure analysis unit 14.

  In step S4, the sentence structure analysis unit 14 analyzes the sentence structure such as case analysis and dependency analysis of each sentence that has been generalized based on the sentence structure knowledge data. Move. As a specific example, in the process of step S4, as shown in FIG. 8, the relationship between terms in each sentence is determined, and the causal relationship in each sentence is extracted. In FIG. 8, the causal relationship is expressed by drawing an arrow from the event corresponding to the cause to the event corresponding to the result. In step S4, data obtained as a result of sentence structure analysis (result of sentence structure analysis) is sent to the inference unit 23.

  In step S 5, the inference unit 23 receives the sentence structure analysis result (causal relation data) sent from the sentence structure analysis unit 14 and the causal network data stored in the causal network data storage unit 20. Similarity search is performed using the causal path on the causal network with the most overlap. Then, the case obtained as a result of the similarity search is output to a display unit (not shown), and the flow ends.

  Next, an operation flow in the inference unit 23 will be described with reference to FIGS. 9 to 11B. FIG. 9 is a flowchart showing an operation flow in the inference unit 23 of the causal inference apparatus 1. FIG. 10 is a diagram illustrating an example of a causal path obtained as a result of matching in the matching unit 31. FIG. 11A shows a part of the result of extracting a sentence including a heavy term from a certain case. FIG. 11B is a diagram showing a part of the causal relationship obtained from a certain case.

  First, in step S <b> 51, the matching unit 31 receives a sentence structure analysis result from the sentence structure analysis unit 14. Subsequently, in step S52, the matching unit 31 superimposes the result of the sentence structure analysis on the data of the causal network while inserting the null node in the non-matching place without changing the order of the text data (matching is performed). . Specifically, the text data as shown in FIG. 8 obtained as a result of the analysis of the sentence structure is matched with the data of the causal network in order according to the arrangement order, and the data on the causal network having the best causal structure. Select (generate) a causal path. Note that the causal path on the causal network where the causal structure most closely matches here is the last match from the node where the text data obtained as a result of the structural analysis and the data of the causal network match first. Represents the path to the selected node. A score on the causal network is assigned to the causal path. Further, in the present embodiment, not only the causal path on the causal network with the most matching causal structure but also a causal path having a causal structure similarity of a predetermined height or higher is selected (generated). Good. Note that the similarity described here refers to a score based on the ratio that matches the causal path, and is obtained in the same manner as the similarity described above. The predetermined height is arbitrarily set.

  In the present embodiment, a configuration has been shown in which the causal path on the causal network with the most overlap is selected, but the present invention is not necessarily limited thereto. For example, the text data as shown in FIG. 8 obtained as a result of the analysis of the sentence structure is matched with the data of the causal network while inserting null nodes in the order that does not match in order according to the arrangement order. The nodes and causal network data derived from the text data obtained as a result of the analysis of the sentence structure as shown in FIG. 10 by replacing the null node inserted in the path between the locations with the causal network nodes. The structure which acquires the causal path produced | generated from the node originating in may be sufficient. In the causal path generated from the node derived from the text data obtained as a result of the sentence structure analysis and the node derived from the data of the causal network, the nodes derived from the text data obtained as a result of the sentence structure analysis In addition to the first node to the last node, the nodes between the first node and the last node are inserted into the nodes from the causal network data as necessary. . Referring to a specific example in FIG. 10, shaded locations in the diagram are nodes derived from causal network data, and unshaded locations are derived from text data obtained as a result of sentence structure analysis. It is a node. The causal path is given a score on the causal network, and an arc between nodes derived from text data obtained as a result of structural analysis is given a maximum score of 1.0.

  In the present embodiment, it is assumed that the similarity between the text data obtained as a result of the sentence structure analysis and the data of the causal network is equal to or higher than a predetermined height, in addition to the best match. The configuration may be such that a plurality of causal path candidates are generated. Note that the similarity described here refers to a score based on the ratio that matches the causal path, and is obtained in the same manner as the similarity described above. In addition, the predetermined height here is arbitrarily set.

  In step S <b> 53, the similarity search unit 32 selects a causal relationship data group including a causal relationship similar to the causal path selected by the matching unit 31 from the causal relationship data group stored in the causal relationship data storage unit 22. . As an example, the causal relationship as shown in FIG. 11B and the causal path shown in FIG. 10 obtained based on the result of extracting a sentence including a heavy term from a case as shown in FIG. The similarity to is explained. Among the causal relationships shown in FIG. 11B, the causal relationships that coincide with the paths on the causal network shown in FIG. 10 are “crack / crack occurred due to“ material degradation ””, “ This is a causal relationship that “content leak” occurred due to “crack” and “fire / explosion occurred due to“ flammable leak ””. Then, the similarity search unit 32 adds up the scores given to the matched paths as described above, and uses it as an index of the strength of similarity between the causal path and the causal relation data group for each case. In the similarity search unit 32, the strength of similarity is obtained for each case in this way, and the causal relationship data group with the highest score or the higher-ranked causal relationship data group with the highest score is used as the above-mentioned similar causal relationship. Is selected as a causal relationship data group including

  Subsequently, the similarity search unit 32 identifies a case corresponding to the causal relationship data group obtained as a result of the selection based on an ID for identifying each case assigned to each case data group, and corresponds. Read the case from the case database. And the similarity search part 32 outputs the read example to the display part which is not shown in figure as a search result, and complete | finishes a flow.

  Next, the effect obtained by the present invention will be described with reference to FIGS. FIG. 12 is a diagram illustrating an example of a result of a similarity search using a conventional search method. FIG. 13 is a diagram showing an example of the result of the similarity search by the search method of the present invention. Here, for the purpose of investigating the cause of failure, a case where a search is performed from a database containing past failure cases will be described as an example.

  First, FIG. 12 is used to show an example in which a search is performed using a conventional search method in which a search is performed based on the surface layer similarity of character strings representing a target phenomenon. As shown in FIG. 12, in the conventional search method, the search is performed based on the surface similarity with the search character string. An example of phenomenon B including common terms such as “stop”, “leakage”, and “ignition” is obtained as a search result. However, between the phenomenon A and the phenomenon B, the cause of the result of “ignition” is different from “deterioration of piping” and “valve clogging due to an increase in oil viscosity”, respectively. When a search is performed, it is impossible to search for a case suitable for investigating the cause of the failure from past failure cases.

  On the other hand, according to the search method of the present invention, a causal relationship is extracted from the search character string, and a case similar to the causal relationship is searched based on the extracted causal relationship. As shown in Fig. 2, when a sentence representing phenomenon A is searched as a search character string, a case of phenomenon C having a common causal relationship "crack / outflow caused by" material degradation "" is obtained as a search result. Between the phenomena A and C, the cause of the result of “ignition” is “deterioration of the piping” and “deterioration of the tank”, respectively. Is obtained. Accordingly, it can be estimated that the cause of “ignition” is deterioration of the material. As described above, according to the present invention, it is possible to search from past failure cases for cases suitable for investigating the cause of failure. That is, the search accuracy can be further improved in the case search for finding the cause that leads to the target phenomenon.

  As an application example of the causal reasoning apparatus 1 of the present invention, for the purpose of investigating the cause of failure, a search is performed from a database containing past failure cases. That is, using a character string representing an observable phenomenon about the failure that has occurred as a search character string, the causal reasoning apparatus 1 is used to search for a case having similar causal relationships from a database containing past failure cases. It is possible. In addition, based on the contents of customer inquiries to the call center about breakdowns, search for similar cases of the causes of breakdowns from a database of past breakdown cases, or based on the contents of physical measurement data Applying to problem-solving processes that require analogy, such as inferring current and future illnesses by searching for similar cases from a database of past disease cases Is possible.

  Finally, each block of the causal reasoning apparatus 1 may be configured by hardware logic, or may be realized by software using a CPU as follows.

  That is, the causal inference apparatus 1 includes a CPU (central processing unit) that executes instructions of a control program that realizes each function, a ROM (read only memory) that stores the program, and a RAM (random access memory) that expands the program. And a storage device (recording medium) such as a memory for storing the program and various data. An object of the present invention is to provide a recording medium in which a program code (execution format program, intermediate code program, source program) of a control program of the causal reasoning apparatus 1 which is software that realizes the above-described functions is recorded so as to be readable by a computer. This can also be achieved by supplying the causal reasoning apparatus 1 and reading and executing the program code recorded on the recording medium by the computer (or CPU or MPU).

  Examples of the recording medium include a tape system such as a magnetic tape and a cassette tape, a magnetic disk such as a floppy (registered trademark) disk / hard disk, and an optical disk such as a CD-ROM / MO / MD / DVD / CD-R. Card system such as IC card, IC card (including memory card) / optical card, or semiconductor memory system such as mask ROM / EPROM / EEPROM / flash ROM.

  Moreover, the causal inference apparatus 1 may be configured to be connectable to a communication network, and the program code may be supplied via the communication network. The communication network is not particularly limited. For example, the Internet, intranet, extranet, LAN, ISDN, VAN, CATV communication network, virtual private network, telephone line network, mobile communication network, satellite communication. A net or the like is available. Also, the transmission medium constituting the communication network is not particularly limited. For example, even in the case of wired such as IEEE 1394, USB, power line carrier, cable TV line, telephone line, ADSL line, etc., infrared rays such as IrDA and remote control, Bluetooth ( (Registered trademark), 802.11 wireless, HDR, mobile phone network, satellite line, terrestrial digital network, and the like can also be used. The present invention can also be realized in the form of a computer data signal embedded in a carrier wave in which the program code is embodied by electronic transmission.

  The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the claims, and the technical means disclosed in different embodiments can be appropriately combined. Such embodiments are also included in the technical scope of the present invention.

  As described above, the causal reasoning apparatus, the control program, and the control method thereof according to the present invention make it possible to further improve the search accuracy in the case search for obtaining the cause associated with the target phenomenon. Therefore, the present invention can be suitably used in an industrial field related to an apparatus that obtains a cause that leads to a target phenomenon by a case search.

It is a functional block diagram which shows one Embodiment of the causal inference apparatus in this invention. (A) is a figure which shows an example of the result of having extracted the sentence containing a heavy term from a some example, (b) is a figure which shows an example of the causal relationship obtained from the some example. (A) is a figure which shows an example of the causal network in this invention, (b) is a figure which shows an example of matching with causal relation and the strength of causal relation. It is a flowchart which shows the operation | movement flow in the said causal inference apparatus. It is a figure which shows an example of the text input into the input part of the said causal inference apparatus. It is a figure which shows an example of selection of the sentence containing the important word in this invention. It is a figure which shows an example of the generalization of the term in this invention. It is a figure which shows an example of the analysis result of the sentence structure in this invention. It is a flowchart which shows the operation | movement flow in the reasoning part of the said causal reasoning apparatus. It is a figure which shows an example of the causal path obtained as a result of the matching in the matching part of the said causal reasoning apparatus. (A) is a diagram showing an excerpt of a part of the result of extracting a sentence containing a heavy term from a case. (B) is an excerpt of a part of the causal relationship obtained from a case. FIG. It is a figure which shows an example of the result of the similar search by the conventional search method. It is a figure which shows an example of the result of the similarity search by the search method of this invention.

Explanation of symbols

1 Causal Reasoning Device 11 Input unit (text data acquisition means)
12 1st important word extraction part 13 1st generalization process part 14 Sentence structure analysis part (1st causal relationship extraction means)
15 Case data acquisition unit (Case data acquisition means)
16 second important word extraction unit 17 second generalization processing unit 18 causal relationship extraction unit (second causal relationship extraction means)
19 Causal network creation unit (causal network creation means)
20 Causal network data storage unit 21 Causal relationship data collection unit (causal relationship data collection means)
22 Causal relationship data storage unit (causal relationship data storage unit)
23 Inference part 31 Matching part (inference means)
32 Similarity search part (similarity search means)

Claims (6)

Text data acquisition means for acquiring text data describing the phenomenon of interest;
First causal relationship extraction means for extracting a causal relationship between words in a sentence included in the sentence by analyzing the sentence structure in the accepted sentence;
Case data acquisition means for acquiring case data from a database storing a plurality of case data that are sentences written in advance about the phenomenon;
A second causal relationship extracting means for extracting a causal relationship between words in a sentence included in the sentence in the sentence by analyzing a sentence structure in the sentence of the acquired case;
The causal network data obtained by connecting each of the plurality of causal relationships extracted from the plurality of cases by the second causal relationship extracting means according to the overlap between the words and phrases constituting the causal relationship, and representing the data of the causal network. A causal network creation means to create,
Based on the causal relation data extracted by the first causal relation extracting means and the data of the causal network, it is inferred which path on the causal network the causal relation is similar to, and based on the inferred path A causal inference apparatus comprising an inference means for generating a causal path. Causal relation data collecting means for collecting the data of the causal relation extracted by the second causal relation extracting means for each case;
A causal relationship data storage unit that stores causal relationship data groups that are data of causal relationships collected for each case separately for each case;
Selecting a causal relation data group similar to the causal path generated by the inference means from the causal relation data storage unit, and further comprising a similar search means for reading out a case corresponding to the selected causal relation data group from the database. The causal inference apparatus according to claim 1, wherein the causal inference apparatus is characterized by the following.   2. The causal reasoning apparatus according to claim 1, wherein the causal path is a path on the causal network corresponding to a place where the causal relation extracted by the first causal relation extracting unit is similar to the causal network.   The causal path is a path obtained by connecting the causal relationship to a path on the causal network corresponding to a location where the causal relationship extracted by the first causal relationship extracting unit is similar to the causal network. The causal reasoning apparatus according to claim 1. A sentence data acquisition step of acquiring sentence data described about the target phenomenon by the sentence data acquisition means;
A first causal relationship extracting step of extracting a causal relationship between words in a sentence included in the sentence by analyzing a sentence structure in the sentence received in the sentence data acquiring step by a first causal relation extracting unit;
A case data acquisition step for acquiring case data from a database storing a plurality of case data, which are sentences preliminarily described about the phenomenon, by case data acquisition means;
The second causal relationship extraction unit extracts the causal relationship between words in the sentence included in the sentence in the case by analyzing the sentence structure in the sentence of the case acquired in the case data acquisition step by the second causal relation extracting unit. Steps,
Each of a plurality of causal relationships extracted from a plurality of cases in the second causal relationship extraction step by the causal network creation means is connected in accordance with the overlap between words constituting the causal relationship and represented in a network form. A causal network creation step for creating causal network data;
Based on the causal relation data extracted in the first causal relation extracting step and the data of the causal network, the reasoning means infers which path on the causal network is similar to the causal network and inferred from the reasoning means. A causal inference apparatus control method comprising: an inference step for generating a causal path based on the path. On the computer,
A sentence data acquisition step for acquiring sentence data describing the phenomenon of interest;
A first causal relationship extracting step of extracting a causal relationship between words in the sentence included in the sentence by analyzing the sentence structure in the accepted sentence;
A case data acquisition step of acquiring case data from a database storing a plurality of case data which are sentences written in advance about the phenomenon;
A second causal relationship extracting step of extracting a causal relationship between words in the sentence included in the sentence in the case by analyzing the sentence structure in the sentence of the acquired case;
The causal network data obtained by connecting each of the plurality of causal relationships extracted from the plurality of cases by the second causal relationship extracting means according to the overlap between the words and phrases constituting the causal relationship, and representing the data of the causal network. A causal network creation step to create,
Based on the causal relation data extracted by the first causal relation extracting means and the data of the causal network, it is inferred which path on the causal network the causal relation is similar to, and based on the inferred path A control program for a causal inference apparatus that operates an inference step for generating a causal path.