JP5160120B2 - Information search apparatus, information search method, and information search program - Google Patents

Description

The present invention relates to an information search apparatus, an information search method, and an information search program that output a sentence similar to an input sentence.

  Conventionally, various search techniques have been provided and widely used for searching data. For example, an Internet search engine searches for documents that match a user's search conditions, and ranks the searched documents according to the degree of similarity.

  In this document search, a document is searched based on a keyword input by a user or a keyword extracted from an input sentence. However, in natural language, the same meaning as these keywords can be expressed in different words.

  For example, when a search request for a document including a search request sentence “send reading” is made from the user, the search request sentence “send reading” is rephrased “read and send”. A document including the sentence (hereinafter referred to as sentence A) is not detected. For this reason, the user may search for a document using “reading” and “send” as search keywords in the search request sentence “send reading”. In this case, a document including the sentence “giving reading and executing transmission” (sentence A) is surely detected, but “generating a reading from a character string and executing voice transmission” (hereinafter referred to as “transmission”). A document including a sentence whose meaning is completely different from “send reading” is also detected. In this case, since a document unintended by the user is also detected, it is difficult to specify a document desired by the user. As described above, there is a problem that it is difficult to distinguish the sentence A and the sentence B in the search by the keyword.

  Hereinafter, the above-described “transmission of reading” is referred to as “paraphrase”, which is the same semantic content and different expression, such as “add reading to character string and execute transmission”.

  Then, generating a candidate for another expression (= paraphrase expression) having the same semantic content from a certain expression is called a paraphrase expression generation technique. As a method for generating this paraphrase expression, there is a technique in which a certain expression is partially replaced and another expression is generated based on a replacement rule stored in advance. For example, in the technique described in Patent Document 1, a method of generating a replacement candidate by applying a plurality of replacement rules is proposed. Thereby, a complicated paraphrase can be realized. It is also conceivable to apply this paraphrase expression generation method to search. That is, when a search target sentence is generated from a search request sentence using a paraphrase expression generation method, and it is determined that the search target sentence is similar to the search request sentence, the search target sentence is used. A method of performing a search can be considered.

JP 2003-76687 A

  However, although the technique described in Patent Document 1 changes the wording according to the rewrite rule, it does not perform a complicated paraphrase that changes the number of independent words. For this reason, there is a problem that a paraphrase between “send reading” and “apply reading to a character string and execute transmission” cannot be performed.

  The present invention has been made in view of the above, and an object of the present invention is to provide an information search apparatus, an information search method, and an information search program that can acquire a properly paraphrased (replaced) sentence.

In order to solve the above-described problems and achieve the object, an information search apparatus according to the present invention includes an input unit that receives input of an input sentence, a search target sentence that is to be searched by the input sentence, and the search target sentence. , A sentence storage means for storing a self-supporting word for each phrase and a syntax tree that holds a dependency structure between self-supporting words with the self-supporting word as a tree-structured node, and parsing the input sentence A syntax tree generating means for generating a syntax tree that holds a dependency structure between independent words of the input sentence, a first partial syntax tree pattern indicating a dependency structure between a plurality of independent words, and A second partial syntax tree pattern indicating a dependency structure between independent words that is a paraphrase of one partial syntax tree pattern , wherein the number of nodes is smaller than that of the first partial syntax tree pattern; Before the first partial syntax tree pattern A syntax tree pattern storage means in association with replacement condition for replacing the second partial parse tree pattern, a partial parse tree contained in the search object text stored in the text storage means, the syntax Determining means for determining whether or not the replacement condition stored in the tree pattern storage means matches; and when determining that the replacement condition matches, the first partial syntax tree pattern associated with the replacement condition; Replacement means for replacing the partial syntax tree included in the matching search target sentence with the second partial syntax tree pattern associated with the replacement condition, and the input generated by the syntax tree generation means When the similarity determination unit that determines the similarity between the syntax tree of a sentence and the syntax tree of the search target sentence replaced by the replacement unit, and the determined similarity satisfies a predetermined condition, The search subject sentence before being replaced by serial substitution means, characterized in that it comprises an output means for outputting the retrieval result.

The information search method of the present invention includes an input step for receiving input of an input sentence, a search target sentence to be searched by the input sentence, an independent word for each phrase in the search target sentence, and the independent word A syntax tree that holds a dependency structure between independent words having a tree structure as a node, and a sentence storage step that associates and stores the syntax tree in the storage unit, and the input sentence is parsed and the independent word of the input sentence A syntax tree generation step of generating a syntax tree that retains a dependency structure between them, and a partial syntax tree included in the search target sentence stored in the storage means includes the first partial syntax tree pattern as the first partial syntax tree pattern. 2 is a partial syntax tree indicating a dependency structure between a plurality of independent words when it is determined that the replacement condition matches with the replacement condition to be replaced with the partial syntax tree pattern of FIG. The above replacement clause The partial syntax tree included in the search target sentence that matches the first partial syntax tree pattern associated with the first partial syntax tree pattern is associated with the first partial syntax tree pattern and the first partial syntax tree pattern. substituted replaced by a partial parse tree pattern dependency second partial parse tree pattern fewer of said node from partial parse tree wherein a pattern first partial parse tree pattern showing the structure between content words a paraphrase And a similarity determination step for determining a similarity between the syntax tree of the input sentence generated by the syntax tree generation step and the syntax tree of the search target sentence replaced by the replacement step. An output step of outputting, as a search result, the search target sentence before being replaced by the replacement step when the similarity is equal to or greater than a predetermined reference value. And butterflies.

The information search program of the present invention includes an input step for receiving input of an input sentence, a search target sentence to be searched by the input sentence, an independent word for each phrase in the search target sentence, and the independent word A syntax tree that holds a dependency structure between independent words having a tree structure as a node, and a sentence storage step that associates and stores the syntax tree in the storage unit, and the input sentence is parsed and the independent word of the input sentence A syntax tree generation step of generating a syntax tree that retains a dependency structure between them, and a partial syntax tree included in the search target sentence stored in the storage means includes the first partial syntax tree pattern as the first partial syntax tree pattern. 2 is a partial syntax tree indicating a dependency structure between a plurality of independent words when it is determined that the replacement condition matches with the replacement condition to be replaced with the partial syntax tree pattern of FIG. Said The partial syntax tree included in the search target sentence that matches the first partial syntax tree pattern associated with the conversion condition is associated with the first partial syntax tree pattern and the in the second partial parse tree pattern number is less of a first partial parse tree pattern of paraphrase become independent words between in charge said receiving a partial parse tree pattern showing the structure first partial parse tree pattern than the node A replacement step, a similarity determination step for determining a similarity between the syntax tree of the input sentence generated by the syntax tree generation step and the syntax tree of the search target sentence replaced by the replacement step; An output step for outputting the search target sentence before being replaced by the replacement step as a search result when the determined similarity satisfies a predetermined condition; To be executed by the over data.

  According to the present invention, it is possible to specify a search target sentence similar to the input sentence with high accuracy.

  Exemplary embodiments of an information search apparatus, an information search method, and an information search program according to the present invention will be described below in detail with reference to the accompanying drawings.

(First embodiment)
As shown in FIG. 1, the information search apparatus 100 includes a search target sentence storage unit 101, a replacement rule storage unit 102, a replacement order storage unit 103, a replacement history storage unit 104, an input unit 105, and a syntax analysis unit. 106, a phrase specifying unit 111, a determining unit 107, a replacing unit 108, a similarity determining unit 109, a display processing unit 110, and a search request sentence matching phrase storage unit 112.

  The search target sentence storage unit 101 stores in advance a syntax analysis result for a search target sentence that is to be searched by a search request sentence input by a user. As shown in FIG. 2, in the search target sentence management table, the relationship between the sentence ID, the search target sentence, the phrase ID, the phrase independent word, the part of speech of the independent word, the parent ID, and the next phrase. And the matching phrase ID are stored in association with each other.

  The sentence ID is an ID assigned to each search target sentence. The phrase ID is an ID assigned to each phrase of the search target sentence. As the relationship, the relationship with the next clause is stored, and for example, “no”, “=”, “parallel”, “adjacent”, and the like. Note that the matching phrase ID is left blank in the initial state, and is used when processing to be described later is performed. The parent ID stores a phrase ID indicating the parent phrase. That is, in the search target sentence management table, the structure of the syntax tree that is the result of the syntax analysis of the search target sentence is stored by this parent ID. Next, the structure of the syntax tree of the sentence ID “1” indicated by reference numeral 201 is shown.

  The syntax tree of the sentence ID “1” shown in FIG. 3 is a syntax tree for the sentence ID “1”, and the syntax tree structure is based on the parent ID that is the field information of the search target sentence management table shown in FIG. Is realized. As shown in FIG. 3, “Grant (Phrase ID“ 2 ”)” and “Transmission (Phrase ID“ 3 ”)” are associated with “Execution (Phrase ID“ 4 ”)”. . Furthermore, “reading (sentence ID“ 1 ”)” is associated with “giving (sentence ID“ 2 ”)”. In the processing described later, the paraphrase processing is performed using such a syntax tree structure.

  The replacement rule storage unit 102 stores replacement rules used for paraphrases performed by a replacement unit 108 described later. According to the replacement rule, the paraphrase of the search target sentence is performed. As shown in FIG. 4, the replacement rule management table holds a classification ID, a rule ID, a target pattern, a replacement pattern, a dependency condition, a part of speech condition, and a matching condition in association with each other. The classification means a predetermined group that becomes a sentence break. In this embodiment, the classification includes noun phrases, verb phrases, case sharing, and the like.

  The classification ID is an ID assigned to each classification described above. The rule ID is an ID assigned to each replacement rule set for each classification. For those with a smaller number of rule IDs, a replacement rule with a complicated dependency structure is set. For those with a larger number of rule IDs, a replacement rule with a simple dependency structure is set. To be there. In other words, if the search target sentence matches a complicated replacement rule, the replacement associated with the replacement rule is performed. If the search target sentence does not match the complicated replacement rule, it is determined whether or not the simple replacement rule matches. If it matches the simple replacement rule, the replacement associated with the replacement rule is performed. Thus, by determining whether or not the matching rules are matched in order from the complicated replacement rule, the accuracy of replacing the syntax tree can be improved.

  The target pattern is a syntax tree pattern indicating a dependency structure between independent words applied to a search target sentence. The replacement pattern is a pattern of a syntax tree after replacement processing executed when a partial syntax tree included in the search target sentence matches the target pattern and satisfies a replacement condition described later. “-” In the target pattern and the replacement pattern indicates that a dependency relationship exists between two words. As shown in FIG. 4, the replacement pattern has a smaller number of nodes than the target pattern. Note that the syntax tree of the target pattern and the syntax tree of the replacement pattern are different in terms of the number of nodes but are semantically different. Thus, the paraphrase relationship is established even if the search target sentence is replaced using the plurality of patterns.

  The example shown in FIG. 5 is an example showing the syntax tree of the target pattern of the record 401 in FIG. 4 and the syntax tree of the replacement pattern, and the target included in the target sentence when a predetermined condition is satisfied. The syntax tree of the pattern is replaced with the syntax tree of the replacement pattern.

  Returning to FIG. 4, the dependency condition, part-of-speech condition, and matching condition in the replacement rule management table are set as replacement conditions for leaving the independent words of the search request sentence and replacing the target pattern with the replacement pattern in the search target sentence. Has been. When all these replacement conditions are satisfied, the replacement process is performed on the search target sentence. The dependency condition indicates a condition for dependency (relationship) between phrases. For example, when the record 401 is used, the determination unit 107 described later determines that the condition is satisfied when the dependency between the clauses B and A is “no” or “adjacent”. The part-of-speech condition indicates the part-of-speech condition of the independent word of each phrase. The matching condition is set on the condition that whether or not the independent word of each clause of the search target sentence matches the independent word of the phrase of the search request sentence. “Disagree” indicates that it does not match the independent word of the phrase of the search request sentence, and “Match” indicates that it matches the independent word of the phrase of the search request sentence.

  That is, in the replacement rule of the record 401 with the classification ID = 2 and the rule ID = 1, the noun B is related to the noun A, and the noun A includes a pattern related to the word (part of speech is arbitrary) C in the search target sentence. In this case, the determination unit 107 described later indicates that the relationship between the noun B and the noun A is “no” or “neighbor” between these phrases, and the noun B matches the phrase of the search request sentence. If it does not match the clause of the search request sentence, it is determined that the replacement condition is satisfied. Then, the replacement unit 108 described later replaces the noun B in the search target sentence with the noun A, the noun A replacing the pattern related to the word C, and the noun B replacing the pattern related to the word C.

  The replacement order storage unit 103 stores a replacement order management table that holds the calling order of classifications for which the above-described replacement rules are set. As shown in FIG. 6, the replacement order management table holds a classification ID and a rule classification in association with each other. And in the information search device 100 concerning this Embodiment, suppose that a replacement process is performed in an order from the record on the said replacement order management table. In other words, processing is first performed using the verb phrase replacement rule, then processing is performed using the noun phrase replacement rule, and finally processing is performed using the case sharing replacement rule.

  The input unit 105 receives an input of a search request sentence from a user and performs an input process of the input search request sentence.

  The syntax analysis unit 106 performs syntax analysis on the search request sentence input by the input unit 105, and generates a syntax tree indicating a dependency structure between independent words for each clause of the search request sentence. Also, the part-of-speech determination process for the independent words of each phrase is performed. Next, a description will be given of a result of parsing performed by the parsing unit 106 when the input unit 105 performs input processing of a search request sentence “send reading”.

  As shown in FIG. 7, the parsing unit 106 indicates that the search request sentence includes two clauses, and that there is a dependency relationship between the noun “read” and the verb “transmit”. To analyze. The syntax analysis unit 106 further analyzes the dependency relationship as “O”.

  The phrase specifying unit 111 specifies a phrase including an independent word that matches (or matches a word registered separately as a similar word) between the search target sentence and the search request sentence. Then, when the phrase specifying unit 111 specifies a phrase including the matching independent word, the phrase ID indicating the phrase of the search request sentence including the matched independent word is used as the matching phrase ID to search the search target sentence storage unit 101. Store in the target statement management table. The detailed processing procedure will be described later.

  In addition, when the phrase specifying unit 111 specifies a phrase including the matching independent word, the phrase ID indicating the phrase of the search target sentence including the matched independent word is used as the matching phrase ID of the search request sentence. The search request sentence matching phrase management table of the phrase storage unit 112 is stored.

  The search request sentence matching phrase storage unit 112 stores a search request sentence matching phrase management table. The search request sentence matching phrase management table holds, for each phrase of the search request sentence, a matching phrase of the search target sentence in association with it. The table structure of the search request sentence matching phrase management table will be described later.

  The determination unit 107 uses, in the classification order stored in the replacement order storage unit 103, the clauses identified as independent words in the phrase specification unit 111 and the replacement rules of the classification stored in the replacement rule storage unit 102. Thus, it is determined whether or not the replacement condition for replacing the target pattern with the replacement pattern is satisfied. The detailed processing procedure will be described later.

  The replacement unit 108 matches the target pattern stored in the replacement rule storage unit 102 among the partial syntax trees included in the search target sentence when the determination unit 107 determines that the replacement condition is satisfied. The partial syntax tree to be replaced is replaced with the partial syntax tree indicated as the replacement pattern. Then, the replacement unit 108 stores the search target sentence after the replacement process in the replacement history storage unit 104 described later. The detailed processing procedure will be described later.

  In this embodiment, the determination unit 107 and the replacement unit 108 perform processing on all search target sentences stored in the search target sentence storage unit 101. However, the processing is not limited to such processing. Absent. For example, after the search target sentence that is likely to be similar to the search request sentence is extracted in advance, the processing by the determination unit 107 and the replacement unit 108 may be performed.

  The replacement history storage unit 104 stores the search target sentence after the replacement process performed by the replacement unit 108 for each processing step. The stored data and the like will be described later.

  The similarity determination unit 109 determines the similarity between the search target sentence after replacement processing stored in the replacement history storage unit 104 and the search request sentence input by the user. A detailed similarity determination method will be described later.

  The display processing unit 110 performs a display process on the search target sentence together with the determination result by the similarity determination unit 109, that is, a process for outputting the search target sentence to a display unit described later. Further, the display processing unit 110 performs display processing of the replacement result for each classification stored in the replacement history storage unit 104 when the search target sentence is displayed.

  Next, an overall processing procedure performed by the information search apparatus 100 shown in FIG. 1 will be described with reference to FIG.

  First, the input unit 105 performs a search request sentence input process (step S801). Next, the syntax analysis unit 106 performs syntax analysis on the search target sentence that has been subjected to the input process, and generates a syntax tree of the search target sentence (step S802).

  Then, the phrase specifying unit 111 specifies a phrase including an independent word that matches between the search target sentence and the search request sentence (step S803). Then, when the phrase specifying unit 111 specifies a phrase including the matching independent word, the phrase ID indicating the phrase of the search request sentence including the matched independent word is used as the matching phrase ID to search the search target sentence storage unit 101. Store in the target statement management table.

  Next, the determination unit 107 and the replacement unit 108 perform replacement processing of the partial syntax tree included in the search target sentence (step S804).

  Then, the similarity determination unit 109 determines the similarity between the search target sentence replaced in step S804 and the search request sentence (step S805).

  Then, the display processing unit 110 performs search target sentence display processing based on the determination result by the similarity determination unit 109 (step S806).

  Next, a processing procedure for specifying a phrase including an independent word that matches between the search target sentence and the search request sentence in the phrase specifying unit 111 shown in step S803 of FIG. 8 will be described with reference to FIG.

  First, the phrase specifying unit 111 sets ‘1’ as an initial value for the variable i and the variable j (step S <b> 901).

  Next, the phrase specifying unit 111 determines whether or not the independent word of the phrase ID “i” of the search request sentence matches the independent word of the phrase ID “j” of the search target sentence that is the current processing target ( Step S902). If the phrase specifying unit 111 determines that they do not match (step S902: No), the phrase specifying unit 111 does not particularly perform setting processing between these phrase IDs.

  When the phrase specifying unit 111 determines that the independent words match (step S902: Yes), the phrase specifying unit 111 sets “j” as the matching phrase ID of the phrase ID “i” of the search request sentence (step S903). Then, the phrase specifying unit 111 registers the matching phrase ID in the search request sentence matching phrase management table.

  As shown in FIG. 10, the search request sentence matching phrase management table stores a matching phrase ID that is a phrase ID that matches each sentence ID of the search target sentence for each phrase ID of the search request sentence.

  Returning to FIG. 9, the phrase specifying unit 111 sets “i” as the matching phrase ID of the phrase ID “j” of the search target sentence (step S904). Then, the phrase specifying unit 111 registers the matching phrase ID in the search target sentence management table.

  As shown in FIG. 11, the phrase ID of the matched search request sentence is registered in the matched phrase ID associated with the phrase ID of each search target sentence.

  Returning to FIG. 9, the phrase specifying unit 111 determines whether there is a phrase that is not processed in the search target sentence (step S905). If it is determined that there is (Step S905: Yes), “1” is added to the variable j (Step S906), and the processing is performed again from Step S902.

  If the phrase specifying unit 111 determines that all the phrases in the search target sentence have been processed (step S905: No), the phrase specifying unit 111 determines whether there is a phrase that has not been processed in the search request sentence (step S905). S907). If it is determined that there is (step S907: Yes), an initial value “1” is set to the variable j indicating the phrase ID to be processed of the search target sentence, and “1” is added to the variable i (step S908), and again. Processing is performed from step S902.

  Next, when the phrase specifying unit 111 determines that all the phrases in the search request sentence have been processed (step S907: No), it determines whether there is a search target sentence that has not been processed (step S907). S909). When it is determined that there is (step S909: Yes), the phrase specifying unit 111 sets a search target sentence that has not been processed as a processing target (step S910), and performs the processing from step S901 again.

  If the phrase specifying unit 111 determines that processing has been performed on all search target sentences (step S909: No), the processing ends.

  With the above-described processing procedure, the phrase of the search target sentence and the phrase of the search request sentence are associated with each other between phrases including the same independent word.

  Next, the replacement processing procedure of the partial syntax tree included in the search target sentence by the determination unit 107 and the replacement unit 108 shown in step S804 of FIG. 8 will be described with reference to FIG.

  First, the determination unit 107 sets the first phrase (phrase ID = 1) of the first search target sentence (sentence ID = 1) in the search target sentence storage unit 101 as a processing target (step S1201). Note that the determination unit 107 sequentially performs processing from the classification ID “1” according to the order stored in the replacement order storage unit 103.

  Next, the determination unit 107 determines whether or not the syntax tree including the clause to be processed matches the condition of the predetermined replacement rule of the classification (initially the replacement rule of the rule ID “1”) (step S1202). ). In the replacement rule management table shown in FIG. 4, it is assumed that the clause focused as a processing target corresponds to the target pattern A. The determination unit 107 satisfies the replacement rule condition when the dependency structure of the target pattern is established and the dependency condition, the part-of-speech condition, and the matching condition associated with the target pattern are satisfied. Judge that

  If the determination unit 107 determines that the conditions of the replacement rule do not match (step S1202: No), it determines whether there is another replacement rule in the classification (step S1203). If the determination unit 107 determines that there is another replacement rule (step S1203: Yes), the determination unit 107 sets the other replacement rule as a replacement rule to be used next (step S1204). In the replacement rule shown in FIG. 4, for example, there are two replacement rules in the classification ID “1”. Therefore, if the condition does not match the condition of the rule ID “1”, the condition of the rule ID “2” is met. Whether or not to do so will be determined. In this way, it is determined whether or not the rule IDs are matched in order from a small number of rule IDs.

  In the replacement rule management table, the smaller the number of rule IDs, the more complicated the replacement condition, and the larger the number of rule IDs, the simpler the replacement condition is set (for example, classification ID = 3). See replacement rules). As a result, when a complicated replacement condition is matched, a replacement process associated with the complex replacement condition is applied, and a replacement process associated with a simple replacement condition is not applied. By setting the priority order in the replacement rule as described above, the information search apparatus 100 according to the present embodiment can perform more appropriate replacement (paraphrase) for the syntax tree of the search target sentence.

  In addition, when the determination unit 107 determines that the replacement condition set in the replacement rule matches (step S1202: Yes), the replacement unit 108 associates the predetermined replacement rule with the predetermined replacement rule in the search target sentence. A process of replacing the partial syntax tree that matches the specified target pattern with a syntax tree that matches the replacement pattern associated with the predetermined replacement rule is performed (step S1205).

  Then, the replacement unit 108 stores the clauses up to the replaced syntax tree among the clauses not stored in the replacement history storage unit 104 in the replacement history storage unit 104 (step S1206).

  When the determination unit 107 determines that there is no other replacement rule in the classification (step S1203: No) or after the processing of step S1206 ends, the determination unit 107 does not perform processing in the processing target search target sentence. It is determined whether there is a phrase (step S1207). If it is determined that there is a clause that has not been processed (step S1207: Yes), the clause next to the clause that was the processing target is set as the processing target (step S1208). Then, the process is performed again from step S1202.

  When the determination unit 107 determines that the processing has been completed for all the clauses (step S1207: No), the clause that has not been replaced and is not stored in the replacement history storage unit 104 is stored in the replacement history storage unit 104. (Step S1209).

  Next, the determination unit 107 determines whether there is a classification that has not been processed for the search target sentence (step S1210). If the determination unit 107 determines that there is a classification that has not been processed (step S1210: Yes), the next classification is set as a processing target in accordance with the classification order stored in the replacement order storage unit 103 (step S1210). S1211).

  If the determination unit 107 determines that processing has been performed for all classifications (step S1210: No), the determination unit 107 determines whether there is a search target sentence that has not been processed (step S1212). If it is determined that there is a search target sentence that has not been processed (step S1212: Yes), the search target sentence that has not been processed is set as the next processing target (step S1213). Further, the determination unit 107 restarts the process from step S1202 using the first replacement rule of the first classification set in the replacement order storage unit 103.

  If the determination unit 107 determines that there is no search target sentence that has not been processed (step S1212: Yes), the process ends.

  As illustrated in FIG. 13A to FIG. 13D, for each search target sentence, a syntax tree of the search target sentence after being replaced in the classification order stored in the replacement order storage unit 103 is stored. Since the history of replacement for each processing step (classification) is stored in this way, it is possible to specify what replacement has been performed for each classification.

  Next, a procedure for determining the similarity between the search target sentence replaced by the similarity determination unit 109 shown in step S805 of FIG. 8 and the search request sentence will be described with reference to FIG. Note that the processing procedure shown in FIG. 14 is performed for each search target sentence.

  First, the similarity determination unit 109 sets ‘1’ as an initial value to the variable i (step S <b> 1401). Next, the similarity determination unit 109 sets “0” as the initial value for the number of link matches (step S1402). The number of link matches indicates the number of independent words included in the clause and the independent words included in the parent clause of the clause between the search target sentence after the replacement process and the search request sentence. Is. When the number of link matches is equal to or greater than a predetermined reference value, it is determined that the search target sentence and the search request sentence are similar.

  Then, the similarity determination unit 109 refers to the search request sentence matching phrase management table to determine whether or not there is a matching phrase ID with the phrase ID “i” of the search request sentence (step S1403). At this time, when the similarity determination unit 109 determines that there is a matching phrase ID (step S1403: Yes), the matching phrase ID is set to the variable “j” (step S1404).

  Next, the similarity determination unit 109 uses the parent phrase of the phrase ID “j” of the search target sentence after the matching phrase ID of the parent phrase ID of the phrase ID “i” of the search request sentence is replaced by the replacement unit 108. It is determined whether or not it matches the ID (step S1405). If it is determined that they do not match (step S1405: No), no particular processing is performed.

  If the similarity determination unit 109 determines that they match (step S1405: Yes), the similarity determination unit 109 adds 1 to the number of link matches (step S1406).

  Next, the similarity determination unit 109 determines whether there is a phrase that has not been subjected to the above-described processing in the search request sentence (step S1407). If it is determined that there is a phrase (step S1407: Yes), “1” is added to the variable i (step S1408), and the process is restarted from step S1403.

  Also, when the similarity determination unit 109 determines that the above-described processing has been performed on all the clauses (step S1207: No), after setting the current number of link matches as the similarity determination result (step S1409), The process ends.

  By performing the processing described above, the calculated number of link matches is acquired as the similarity determination result for each search target sentence. Then, the display processing unit 110 determines that the similarity search result is similar when the search result is equal to or greater than a predetermined reference value, and performs display processing so that the similarity can be visually recognized.

  In the present embodiment, the predetermined reference value is “1”. Then, when the process shown in FIG. 14 is performed on the search target sentences after the replacement shown in FIGS. 13A to 13D, the similarity determination results of the sentence IDs = 1, 3, and 4 are obtained. Although it is “1” or more, the similarity determination result of the sentence ID = 2 is “0”.

  Therefore, the display processing unit 110 displays that the search target sentences with the sentence ID = 1, 3, and 4 are similar, and displays that the search target sentence with the sentence ID = 2 is dissimilar.

  As shown in FIG. 15, the display processing unit 110 displays “◯” indicating that the sentence IDs = 1, 3, and 4 that are similar are similar, and the sentence ID = 2. Displays “x” indicating dissimilarity. Further, the display processing unit 110 refers to the replacement history storage unit 104 and displays a replacement process for each classification.

  As shown in the example of FIG. 15, the display processing unit 110 passes “provide reading and send” from “giving reading and executing transmission” to the sentence with the sentence ID “1”. And display the replacement process until “send reading”.

  In addition, the display processing unit 110 displays a replacement process from “Generate a reading and execute voice transmission” to “Generate a reading and transmit voice” for the sentence with the sentence ID “2”. I do.

  In addition, the display processing unit 110 sends “reading readings” via “giving and sending readings” from “giving and sending kanji reading information” to the sentence with the sentence ID “3”. Display the replacement process up to “Yes”.

  In addition, the display processing unit 110 performs the process of “adding kanji reading information and executing transmission” to “send kanji reading information and transmitting” to the sentence with the sentence ID “4”. Display the replacement process until “add reading and send”.

  In the present embodiment, the replacement is performed in the classification order (verb phrase, noun phrase, case sharing) stored in the replacement order storage unit 103. The purpose of this is to perform paraphrasing in the order of noun phrases and case sharing after deleting phrases containing unnecessary independent words by first replacing verb phrases.

  Here, in the replacement order storage unit 103, if the calling order of the verb phrase and case sharing is reversed in the classification calling order, the case sharing processing is performed first. In that case, since the unnecessary clause is not deleted in the sentence with the sentence ID “1”, the condition for performing the case sharing process is not satisfied. For this reason, the processing of the sentence to be searched for the sentence ID “1” “provide reading and execute transmission” ends without replacement by the case sharing classification.

  In the information search apparatus 100 according to the above-described embodiment, it is possible to specify a search target sentence similar to a search request sentence with high accuracy. Thereby, for example, it is possible to search for a document including a search target sentence obtained by rephrasing the search request sentence input by the user.

  Also, in the information search apparatus 100 according to the present embodiment, the replacement order storage unit 103 sets the calling order of classifications in an appropriate order so that replacement can be performed with higher accuracy. Yes.

  In the information search apparatus 100 according to the present embodiment, the replacement process for each classification is collectively displayed. On the other hand, if the replacement process is not displayed collectively, the sentence with the sentence ID “4” starts from the search target sentence “Kanji reading information is added and executed”, and the replacement process is “Kanji reading”. “Add information and send”, “Add and send reading information”, and “Add and send reading” are displayed, and then “Send reading” is displayed as the replacement result. become. In this case, since the display becomes complicated, it becomes difficult for the user to easily recognize the process. On the other hand, in the information search apparatus 100 according to the present embodiment, the replacement process with the classification word is displayed as a whole, so that the display of the replacement process is prevented from becoming complicated, and the replacement process is performed by the user. It can be easily recognized.

  As described above, in the information search apparatus 100 according to the present embodiment, the replacement rule is stored in the replacement rule storage unit 102 for each classification, and the replacement process is also performed for each classification. For this reason, it is possible to control the application order of the replacement rules, and to display the replacement process using the replacement rules for each appropriate granularity.

  As a result, it is possible to easily check various paraphrase expression searches that rephrase the search request text and the validity of the search results.

  In the information search apparatus 100 according to the present embodiment, all search target sentences are displayed. However, the display is not limited to such a display format. For example, only search target sentences determined to be similar by the above-described processing are displayed. May be. Furthermore, the present invention can be variously applied without departing from the spirit of the present invention.

  As illustrated in FIG. 16, the information search apparatus 100 includes a ROM 1602 that stores an information output program that performs the above-described processing as a hardware configuration, a CPU 1601 that controls each unit of the apparatus according to a program in the ROM 1602, Random Access Memory (RAM) 1603 for storing various data necessary for controlling the device, a display unit 1605 for displaying the processing results described above, a communication I / F 1604 connected to a network, and an external storage such as a hard disk A device 1606, an input device 1607 used when a user inputs data, and a bus 1608 for connecting each unit are provided. The information search apparatus 100 can be applied to a general computer having the above-described configuration.

  The information output program executed by the information search apparatus 100 according to the above-described embodiment is a file in an installable format or an executable format, and is a CD-ROM, flexible disk (FD), CD-R, DVD (Digital Versatile). The program is recorded on a computer-readable recording medium such as a disk.

  In this case, the information output program is loaded onto the main storage device by being read from the recording medium and executed by each device of the information search device 100, and each unit described in the software configuration is generated on the main storage device. It has come to be.

  The information output program executed by the information search apparatus 100 according to the above-described embodiment is configured to be provided by being stored on a computer connected to a network such as the Internet and downloaded via the network. Also good. In addition, the information output program executed by the information search apparatus 100 according to the above-described embodiment may be provided or distributed via a network such as the Internet.

  Further, the information output program of this embodiment may be configured to be provided by being incorporated in advance in a ROM or the like.

It is a block diagram which shows the structure of the information search device concerning 1st Embodiment. It is the figure which showed the table structure of the search object sentence management table which a search object sentence memory | storage part memorize | stores. It is the figure which showed the tree structure of the syntax tree of sentence ID "1" stored in the search object sentence management table. It is the figure which showed the table structure of the replacement rule management table which a replacement rule memory | storage part hold | maintains. It is explanatory drawing which showed the syntax tree of the object pattern corresponding to the partial syntax tree before substitution, and the syntax tree of the substitution pattern corresponding to the partial syntax tree after substitution. It is the figure which showed the table structure of the replacement order management table. It is the figure which showed the syntax analysis result of the search request sentence “send reading” by the syntax analysis section. It is a flowchart which shows the whole process sequence which an information search device performs. It is a flowchart which shows the process sequence which specifies the phrase containing the independent word which corresponds between the search object sentence and the search request sentence in a phrase specific | specification part. It is the figure which showed the table structure of a search request sentence matching clause management table. It is the figure which showed the search object sentence management table after adding matching clause ID. It is a flowchart which shows the replacement processing procedure of the partial syntax tree contained in the search object sentence by a judgment part and a replacement part. It is the figure which showed the example of the table structure which stored the search object sentence of sentence ID "1" substituted for every classification | category in the replacement history storage part. It is the figure which showed the example of the table structure which stored the search object sentence of sentence ID "2" substituted for every classification | category in the replacement history storage part. It is the figure which showed the example of the table structure which stored the search object sentence of sentence ID "3" substituted for every classification | category in the replacement history storage part. It is the figure which showed the example of the table structure which stored the search object sentence of sentence ID "4" substituted for every classification | category in the replacement history storage part. It is a flowchart which shows the determination process procedure of the similarity between the search object sentence substituted by the similarity determination part, and a search request sentence. It is a figure which shows the example of the search result which a display process part displays. It is a figure which shows the hardware constitutions of an information search device.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Information retrieval apparatus 101 Search object sentence memory | storage part 102 Replacement rule memory | storage part 103 Replacement order memory | storage part 104 Replacement history storage part 105 Input part 106 Syntax analysis part 107 Judgment part 108 Substitution part 109 Similarity determination part 110 Display processing part 111 Clause specification Part 112 Search request sentence matching phrase storage part 1601 CPU
1602 ROM
1603 RAM
1604 Communication I / F
1605 Display unit 1606 External storage device 1607 Input device 1608 Bus

Claims (8)

An input means for receiving an input sentence;
A search target sentence to be searched by the input sentence, an independent word for each phrase in the search target sentence, and a syntax tree that holds a dependency structure between independent words with the independent word as a tree-structured node; , And a sentence storage means for storing them in association with each other,
A syntax tree generating means for parsing the input sentence and generating a syntax tree that retains a dependency structure between independent words of the input sentence;
A first partial syntax tree pattern indicating a dependency structure between a plurality of independent words, and a partial syntax tree pattern indicating a dependency structure between independent words as a paraphrase of the first partial syntax tree pattern , A second partial syntax tree pattern having a smaller number of nodes than one partial syntax tree pattern, and a replacement condition for replacing the first partial syntax tree pattern with the second partial syntax tree pattern, A syntax tree pattern storage means for storing;
Determining means for determining whether a partial syntax tree included in the search target sentence stored in the sentence storage means matches the replacement condition stored in the syntax tree pattern storage means;
When it is determined that the replacement condition is satisfied, the partial syntax tree included in the search target sentence that matches the first partial syntax tree pattern associated with the replacement condition is associated with the replacement condition. Replacement means for replacing with the second partial syntax tree pattern attached;
Similarity determination means for determining similarity between the syntax tree of the input sentence generated by the syntax tree generation means and the syntax tree of the search target sentence replaced by the replacement means;
When the determined similarity satisfies a predetermined condition, an output means for outputting the search target sentence before being replaced by the replacement means as a search result;
An information retrieval apparatus comprising: The syntax tree pattern storage means stores the first partial syntax tree pattern, the second partial syntax tree pattern, and the replacement condition in association with each other for each predetermined group serving as a sentence delimiter. ,
The determination means determines whether the partial syntax tree included in the search target sentence stored in the sentence storage means matches the replacement condition stored in the syntax tree pattern storage means for each predetermined group. To judge,
The information search device according to claim 1. Order storage means for storing the order of determination by the determination means for each predetermined unit;
The determination means includes a partial syntax tree included in the search target sentence stored in the sentence storage means for each predetermined group in the order stored in the order storage means, and the syntax tree pattern storage means includes Determining whether the replacement condition to be stored is met or not;
The information search device according to claim 2. The syntax tree pattern storage means matches, when the simple dependency structure is satisfied, from the complex condition that is determined to match when the complex dependency structure is satisfied for the replacement condition for each predetermined group. Memorize multiple simple conditions to be judged,
The determination means further determines whether or not the replacement condition matches in order from the complex condition to the simple condition;
The information search device according to any one of claims 1 to 3. The output means further outputs the search target sentence in which the syntax tree has been replaced by the replacement means for each predetermined group by the replacement means,
The information search device according to any one of claims 1 to 4, wherein: The syntax tree pattern storage means includes, as the replacement condition, the independent word that the input sentence has in the node included in the partial syntax tree of the search target sentence whose structure matches the first partial syntax tree pattern. One or more of the condition of the position of the node and the condition of the connection relation between the nodes of the partial syntax tree are set,
The information search device according to any one of claims 1 to 5, wherein An input step for accepting input sentences;
A search target sentence to be searched by the input sentence, an independent word for each phrase in the search target sentence, and a syntax tree that holds a dependency structure between independent words with the independent word as a tree-structured node; , And a sentence storage step for storing the same in the storage unit;
A syntax tree generation step of parsing the input sentence to generate a syntax tree that retains a dependency structure between independent words of the input sentence;
It is determined whether or not a partial syntax tree included in the search target sentence stored in the storage unit matches a replacement condition for replacing the first partial syntax tree pattern with the second partial syntax tree pattern. A decision step;
The search, which is a partial syntax tree indicating a dependency structure between a plurality of independent words and matches a first partial syntax tree pattern associated with the replacement condition when it is determined that the replacement condition is satisfied. A portion showing a dependency structure between independent words that is associated with the first partial syntax tree pattern and the partial syntax tree included in the target sentence is a paraphrase of the first partial syntax tree pattern Replacing with a second partial syntax tree pattern that is a syntax tree pattern and has a smaller number of nodes than the first partial syntax tree pattern ;
A similarity determination step for determining a similarity between the syntax tree of the input sentence generated by the syntax tree generation step and the syntax tree of the search target sentence replaced by the replacement step;
An output step of outputting, as a search result, the search target sentence before being replaced by the replacement step when the determined similarity is equal to or greater than a predetermined reference value;
A method for retrieving information, comprising: An input step for accepting input sentences;
A search target sentence to be searched by the input sentence, an independent word for each phrase in the search target sentence, and a syntax tree that holds a dependency structure between independent words with the independent word as a tree-structured node; , And a sentence storage step for storing the same in the storage unit;
A syntax tree generation step of parsing the input sentence to generate a syntax tree that retains a dependency structure between independent words of the input sentence;
It is determined whether or not a partial syntax tree included in the search target sentence stored in the storage unit matches a replacement condition for replacing the first partial syntax tree pattern with the second partial syntax tree pattern. A decision step;
The search, which is a partial syntax tree indicating a dependency structure between a plurality of independent words and matches a first partial syntax tree pattern associated with the replacement condition when it is determined that the replacement condition is satisfied. A portion showing a dependency structure between independent words that is associated with the first partial syntax tree pattern and the partial syntax tree included in the target sentence is a paraphrase of the first partial syntax tree pattern Replacing with a second partial syntax tree pattern that is a syntax tree pattern and has a smaller number of nodes than the first partial syntax tree pattern ;
A similarity determination step for determining a similarity between the syntax tree of the input sentence generated by the syntax tree generation step and the syntax tree of the search target sentence replaced by the replacement step;
When the determined similarity satisfies a predetermined condition, an output step of outputting the search target sentence before being replaced by the replacement step as a search result;
Information retrieval program that causes a computer to execute

Images