JP2007102642A - Information analysis system, information analysis method and information analysis program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To create and output a useful correlation rule among a plurality of pieces of information. <P>SOLUTION: The information analysis system is provided with: a morphological analysis means for performing morphological analysis of input text information; a syntax analysis means; an item creation means for creating a morphological analysis result and/or a syntax analysis result as items for analysis of the correlation rule; an item group creation means for creating one or more item groups using one or more items created by the item creation means; an item group deletion means for deleting item groups having items with semantic inclusive relation as elements by comparing the respective created item groups with one another; an item group computation means for computing appearance frequency of coincidence about the respective item groups; a correlation rule creation means for creating one or more correlation rules based on the computed appearance frequency of coincidence of the respective item groups; and a display means for displaying the respective correlation rules created by the correlation rule creation means. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an information analysis system, an information analysis method, and an information analysis program. For example, the present invention can be applied to an information analysis system that finds and displays useful information from a large amount of data.

  As a data mining technique for finding useful information buried in a large amount of data, for example, association analysis is known. Association analysis is to discover rules related to the connection between events called association rules as knowledge.

  The Priori algorithm disclosed in Non-Patent Document 1 is one of the efficient processing methods of association analysis, and a method for extracting a frequent set greater than the minimum support level has been proposed.

  In association analysis, what an object (hereinafter referred to as an item) to be analyzed is specified according to the purpose. For example, in the case of POS data analysis, a product purchased by a customer may be an item.

  Therefore, the association analysis can also be applied to text mining for natural sentences, and in the case of text mining, it is conceivable to set the item to one of the following.

(Case 1) Word (Case 2) Dependency Relationship Between Words For example, assume that the data is “Mother applied for a specific account advertised on TV but has not been opened”.

  In this case, in case 1, as follows, [mother, television, advertise (is), specific account, apply (is), opens (not)].

  On the other hand, in case 2, since the dependency relationship shown in FIG. 2 exists, [Mother-subscribe (da), TV-advertise (is), advertise (is) -specific account, specific account as follows: -Apply.

  Here, each item has a notation of “relationship source word—relationship destination word”, and the parentheses indicate that the verb is used in a meaning such as negation or continuation (hereinafter referred to as intention information).

  Non-Patent Document 2 discloses a method of discovering important information using the dependency relationship of Case 2, and in that description, an item set is created by combining these items, and the degree of support and confidence A technique for calculating a degree or the like and outputting an association rule with a certain degree of support or certainty is disclosed.

Agrawal, R, Imielinski, T .; , And Swami, A .; “Minning Association Rules between Sets of Items in Large Databases”. Proceedings of ACM SIGMOD-93, PP. 207-216 (1993). Keiko Shimazu, Masahito Monma, Koichi Furukawa, “Discovering Important Information from Call Center Information Using Association Rule Derivation Method”, 17th Annual Conference of Japanese Society for Artificial Intelligence, 1F4-03, 2003 Takayuki Ito, Keisuke Inoue, Satoshi Doi, Yasumasa Tominaga, Yuko Ikebata, “Improvement of Screen Layout Method of Graph Data Using Dynamic Model”, IPSJ Research Reports Graphics and CAD103-2, pp-7-12, 2001 Year

  By the way, when each word appearing in the data is an item (case 1), since each item is a word, the meaning of the obtained association rule is more significant than in the case of dependency relationship (case 2). Difficult to take. On the other hand, in the case of a dependency relationship, each item is in the form of “what is wrong” and is easy to take a meaning.

  Also, in case 1, since the dependency relationship is not taken into account, the item set can be created by simply combining words such as {specific account, apply (da)} without being aware of the main character or object omission. Can be created. On the other hand, in case 2, since the dependency relationship is used, it is not possible to cope with omission of the subject or object specific to the text data.

  For example, in the dependency relationship shown in FIG. 2, “not established” means “(a specific account) is not established”, and the case is omitted. Therefore, in case 2, since there is no dependency relationship between “specific account” and “open (not)”, a dependency relationship including “open (not)” is not created.

  Therefore, an information analysis system that creates and outputs association rules (correlation rules) using dependency relationships and words as items, and can efficiently create and / or output association rules so that they do not overlap. An information analysis method and an information analysis program are provided.

  In order to solve such a problem, an information analysis system according to a first aspect of the present invention is an information analysis system that creates a correlation rule based on each component of a plurality of input text information and outputs a useful correlation rule. (1) morphological analysis means for performing morphological analysis on each text information, (2) syntax analysis means for performing syntax analysis on each text information, and (3) morphological analysis results and / or syntax analysis of each text information. Item creation means for creating a result as an item to be analyzed by the correlation rule, and (4) Item set creation means for creating one or a plurality of item sets using one or a plurality of items created by the item creation means (5) By comparing each item set created by the item set creation means, an item having a semantic inclusion relation is defined as an element. An item set deleting means for deleting the item set, (6) an item set calculating means for calculating the co-occurrence frequency for each item set, and (7) co-occurrence of each item set calculated by the item set calculating means. A correlation rule creating unit that creates one or a plurality of correlation rules based on the appearance frequency, and (8) a display unit that displays each correlation rule created by the correlation rule creating unit.

  An information analysis method according to a second aspect of the present invention is an information analysis method in which a correlation rule is created based on each component of a plurality of input text information and a useful correlation rule is output. A morpheme analysis step for performing morphological analysis on each text information, (2) a syntax analysis step for parsing each text information by a syntax analysis unit, and (3) an item creation unit for each text information An item creation step of creating the morpheme analysis result and / or the syntax analysis result as an item to be analyzed by the correlation rule, and (4) the item set creation means uses one or more items created by the item creation means An item set creation step for creating one or a plurality of item sets, and (5) an item set deletion means each created by the item set creation means An item set deletion step of checking an item set and deleting an item set having items having semantically inclusive relations as elements, and (6) item set calculation means calculates the co-occurrence appearance frequency for each item set An item set calculation step; and (7) a correlation rule creation step in which the correlation rule creation unit creates one or a plurality of correlation rules based on the co-occurrence appearance frequency of each item set calculated by the item set calculation unit; 8) The display means includes a display step of displaying each correlation rule created by the correlation rule creation means.

  An information analysis program according to a third aspect of the present invention is an information analysis program for creating a correlation rule based on each component of a plurality of input text information and outputting a useful correlation rule. Morphological analysis means for performing morphological analysis on text information, (2) Syntax analysis means for performing syntax analysis on each text information, (3) Morphological analysis results and / or syntax analysis results of each text information, Item creation means for creating items to be analyzed, (4) Item set creation means for creating one or more item sets using one or more items created by the item creation means, (5) Item set creation An item that has items that are semantically inclusive as elements by comparing each item set created by the means An item set deleting means for deleting the item set, (6) an item set calculating means for calculating the co-occurrence appearance frequency for each item set, and (7) a co-occurrence appearance frequency of each item set calculated by the item set calculating means. And (8) function as display means for displaying each correlation rule created by the correlation rule creation means.

  According to the present invention, an association rule (correlation rule) is created and output using a dependency relationship and a word as an item, and the creation and / or output of the association rule can be efficiently organized so as not to overlap. it can.

(A) First Embodiment Hereinafter, a first embodiment of an information analysis system, an information analysis method, and an information analysis program of the present invention will be described in detail with reference to the drawings.

  In the present embodiment, a case will be described in which a plurality of items are generated for a text sentence that is a natural sentence, and an association rule in which these items are aggregated is created.

  In addition, the present embodiment extends the technique using the dependency relationship as an item (the technique 2 in the case of the above-described related art) so as to cope with the omission of the subject and object specific to the text data.

  In the extension method, not only dependency relationships but also words such as general nouns, verbs, and adjectives are used as items. Thereby, unlike the case of the above-described conventional technique 1, the dependency relationship is considered, and at the same time, the omission of the subject and the object specific to the text data is supported.

  By using dependency relationships and words as items, there is a semantically overlapping relationship between association rules, but in that case, association rules that are more restrictive are left.

(A-1) Configuration of the First Embodiment FIG. 1 is a block diagram showing the internal configuration of the data analysis device 7A of the first embodiment.

  In FIG. 1, the data analysis device 7 </ b> A of the present embodiment includes an input unit 1, a morpheme analysis unit 2, a syntax analysis unit 3, an item generation unit 4, an association rule extraction unit 5, and a display unit 6.

  Further, the association rule extraction unit 5 includes a control unit 500, a candidate item set generation unit 501, a candidate item set deletion unit 502, a candidate item set calculation unit 503, a rule creation unit 504, and an item set temporary storage unit 505.

  The input unit 1 captures data to be subjected to association analysis, and supplies the captured data to the morpheme analysis unit 2. Further, the input unit 1 takes in “minimum support”, “minimum confidence”, and “maximum size of the item set to be created” for calculating the association rule, and inputs the “minimum support”, “minimum confidence”, “ The maximum size of the item set to be created ”is given to the association extraction unit 5.

  The morpheme analysis unit 2 receives each data fetched by the input unit 1 and performs a morpheme analysis on each data using a predetermined morpheme dictionary or a predetermined rule. This is given to the analysis unit 3.

  The syntax analysis unit 3 receives a morpheme analysis result analyzed by the morpheme analysis unit 2, performs a syntax analysis based on the morpheme analysis result, and extracts a dependency relationship.

  The item generation unit 4 uses the morpheme analysis result by the morpheme analysis unit 2 and the syntax analysis result by the syntax analysis unit 3 to create a plurality of items including words and dependency relationships. The item generation unit 4 gives the created item to the association rule extraction unit 5.

  The association rule extraction unit 5 receives the items created by the item generation unit 4, extracts a necessary item set from the items by a method described later, and creates an association rule. Further, the association rule extraction unit 5 gives the created association rule to the display unit 6.

  When the display unit 6 receives the association rule created by the association rule extraction unit 5, the display unit 6 performs a predetermined output process on the received association rule and outputs it.

  Next, the internal function of the association rule extraction unit 5 will be described with reference to FIG.

  The control unit 500 controls the function of the association rule extraction unit 5.

  The candidate item set generation unit 501 creates item sets of various sizes.

  The candidate item set deletion unit 502 finds and deletes an item set that does not need to be calculated from the semantic relationship between words.

  The candidate item set calculation unit 503 calculates the support level of each item set.

  The rule creation unit 504 generates an association rule and derives an association rule having a certain certainty factor or higher.

  The item set temporary storage unit 505 stores the support level of each item set calculated by the candidate item set calculation unit 503.

(A-2) Operation of First Embodiment Next, the operation of the data analysis device 7 of this embodiment will be described in detail with reference to the drawings.

  FIG. 3 is a flowchart showing the operation of the first embodiment, and the operation will be described along this flowchart.

  When the input unit 1 accepts input of data and takes in the input data, the input unit 1 gives the taken-in data to the morpheme analysis unit 2. The input unit 1 accepts inputs of “minimum support”, “minimum confidence”, and “maximum size of item set to be created”, and these “minimum support”, “minimum confidence”, “creation” When the “maximum size of item set to be performed” is taken in, it is given to the association rule extraction unit 5 (step 100).

  For example, a case where the data shown in FIG. 4 is input data will be described below as an example. Note that “data ID” in FIG. 4 is an identifier for identifying each data.

  Also, for example, “minimum support” = 2 (that is, items appearing in two or more documents are adopted), “minimum certainty” = 0.6, and “maximum size of item set to be created” = 3. . The minimum support level may be the appearance ratio of all documents, but here it is the number of appearances for convenience of explanation.

  When each data input to the input unit 1 is given to the morpheme analysis unit 2, the morpheme analysis unit 2 performs morpheme analysis on each received data (step 110). The morpheme analysis processing by the morpheme analysis unit 2 can use a conventional general morpheme analysis technique, and a detailed description thereof is omitted here.

  Here, FIG. 5 shows the result of the morphological analysis process for “data 1” shown in FIG. In FIG. 5, “data ID” is a data ID of input data, “word ID” is identification information for identifying a morpheme (word) in the input data, and “word” is an analyzed morpheme (word). And “type of morpheme” indicates the part of speech of the word.

  When morpheme analysis is performed on each data by the morpheme analysis unit 2, the syntax analysis unit 3 performs syntax analysis based on the morpheme analysis result of each data (step 120). The parsing process by the parsing unit 3 can use a conventional general parsing technique, and a detailed description thereof is omitted here.

  Here, FIG. 6 shows the result of the parsing process for “data 1 to data 3” shown in FIG. In FIG. 6, “number” is identification information for identifying the dependency relationship in each data.

  The item generation unit 4 creates each dependency relationship as one item (referred to as “dependency item”) for each piece of data based on the syntax analysis result by the syntax analysis unit 3. (Step 130). FIG. 7 shows a result of creating dependency items for “data 1 to data 3” in FIG.

  At this time, the item generation unit 4 deletes the particle information and creates it with the notation “relationship source word + (−) + relationship destination word”. Further, when the relationship source word is a verb, the relationship destination word is a noun, and the particle information is NULL, the item generation unit 4 creates a dependency item in a format in which the relationship destination word and the relationship source word are replaced.

  For example, as indicated by “number 3 of data 2” in FIG. 6, in the case of “advertise (is) + NULL + specific account”, the item generation unit 4 has an item of the form “specific account-advertise (is)”. Is created (see FIG. 7).

  Further, the item generation unit 4 creates a noun, a verb, an adjective, and an adjective verb as an item (this is referred to as a “word item”) based on the morphological analysis result by the morpheme analysis unit 2 (step) 140). FIG. 8 shows a word item creation result for “data 1 to data 3” in FIG. However, the item generation unit 4 does not use nouns such as “things” and “things” as items.

  In the parsing result, when both the source word and the destination word are nouns and the word whose particle information is NULL is a compound noun, the item generation unit 4 determines the meaning of the data only with the source word. Do not register the source word because it may not be represented correctly.

  For example, in “Data 1” in FIG. 4, the word “general account” is a compound noun composed of “general” and “account”, so the item generation unit 4 does not register “general”, which is the dependency source word. . Note that the “account” as the contact word is registered as a word item.

  When the dependency item and the word item of each data are created by the item generation unit 4, the association extraction unit 5 extracts an item set based on the created dependency item and the word item, and creates an association rule ( Step 150).

  FIG. 9 is a flowchart showing detailed processing of the association rule extraction unit 5 in step 150.

  When receiving the dependency item and the word item of each data created by the item generation unit 4, the candidate item set generation unit 501 sets all items received from the item generation unit 4 as item sets of size 1, and sets the item set Registration in the temporary storage unit 505 (step 1000).

  Here, the size of the item set is the number of items (including both dependency items and word items) constituting the item set. For example, the size 1 item set is composed of one item. An item set having a size n (n is a positive integer) is a set composed of n items.

  Then, the candidate item set calculation unit 503 calculates the number of appearances of the size 1 item set for the size 1 item set registered in the item set temporary storage unit 505 (step 1010).

  FIG. 10 shows the number of appearances of a size 1 item set. In FIG. 10, the number of appearances for each item set is associated with the data ID (document ID) in which each item set appears.

  When the number of occurrences of each item set is calculated by the candidate item set calculation unit 503, the candidate item set deletion unit 502 determines that the number of appearances of the item set is less than the “minimum support” received from the input unit 1. Deletes the item set from the item set temporary storage unit 505 (step 1020).

  If all item sets are less than the “minimum support level” and all items are deleted by the candidate item set deletion unit 502 (step 1030), the process ends because there is no association rule that satisfies the condition.

  On the other hand, if there is an association rule that satisfies the condition, the process proceeds to step 1040, and the control unit 500 sets the counter n indicating the size of the item set to 2 in order to process the item set having the size n (2) ( Step 1040).

  When the counter n is set to 2 by the control unit 500, the candidate item set western unit 501 has a size n (here, size 2) based on the item set registered in the item set temporary storage unit 505. An item set is generated (step 1050). Here, for the generation of the item set, for example, the Priori algorithm disclosed in Non-Patent Document 1 may be used.

  When a size 2 item set is generated, the candidate item set deletion unit 502 deletes, for example, an item set including the following subset for each size 2 item set (step 1060).

  For example, an item set such that one item constituting the item set includes all or part of the other item is deleted, such as {W, * -W} or {W, W- *}. However, W is a word and * is an arbitrary word. “−” Represents a dependency relationship.

  For example, an item set of size 2 combining the items “specific account-apply (da)” and “apply (da)” in the eighth and ninth lines in FIG. 10 {specific account-apply (da), apply. (Da)}. In this case, since the “apply (da)” portion of each item is duplicated, the candidate item set deletion unit 502 deletes this item set {specific account—apply (da), apply (da)}. Similarly, the item set {specific account-apply (da), specific account} combining "specific account-apply (da)" and "specific account" in the 8th and 15th lines in FIG. .

  When the item set including the subset is deleted by the candidate item set deletion unit 502, the candidate item set calculation unit 503 calculates the support level (number of appearances) of each item set of size 2, and If the support level is less than the “minimum support level”, the item set is deleted from the item set temporary storage unit 505 (step 1070). FIG. 11 shows an item set of size 2, and deletes the item set in the hatched portion where the number of appearances is less than the “minimum support”.

  Then, when the counter n is equal to the “maximum size of the item set to be created” input from the input unit 1 or when it is less than the “minimum support level” in all the item sets, Proceed to step 1100.

  On the other hand, in other cases, the process proceeds to step 1090, where the control unit 500 adds 1 to the counter n, returns to step 1050, and repeats the process.

  In step 1100, the rule creation unit 504 refers to the item set temporary storage unit 505, and extracts an association rule greater than or equal to “minimum certainty” input to the input unit 1.

  Therefore, first, the rule creation unit 504 selects one item set stored in the item set temporary storage unit 505, and selects the selected item sets as two arbitrary sets (an item set that becomes a condition part and a conclusion part). (Step 1110). The divided sets are denoted as S and 1-S, respectively.

  For example, an item set of size 3 {specific account, apply (da), open (not)} is (1) {specific account, apply (da)}, {open (not)}, (2) { Specific account, open (not)}, {apply (da)}, (3) {apply (da), open (no)}, {specific account}.

  The rule creation unit 504 extracts one set from the divided combinations, and the reliability for the association rule candidates s⇒1-s and 1-s⇒s (one support level for each). / S support degree, 1 support degree / 1-s support degree) (step 1130). Since the method for obtaining the certainty factor is the same as the conventional method for deriving the association rule, a detailed description thereof is omitted.

  If the calculated certainty factor is less than the “minimum certainty factor” input by the input unit 1, the rule creating unit 504 does not adopt the set as an association rule and returns to Step 1120.

  Note that the processing in steps 1130 and 1140 is performed for a set of all combinations obtained by division, and the process returns to step 110 and is performed for all item sets stored in the item set temporary storage unit 505.

  For example, FIG. 12 shows an association rule made from the item set of size 2 shown in FIG. However, those with a certainty factor of less than 0.6 are omitted.

  Also, for example, FIG. 13 shows a size 3 item set, and FIG. 14 shows an association rule made from a size 3 item set.

  In step 1150, a rule with few restrictions on the condition part and the conclusion part of the created association rule is deleted, and the process ends.

  This deletion target is an association rule that simultaneously satisfies the following two conditions.

In the two association rules Ri: Pi⇒Ci (confidence level Cofii), Rj: Pj⇒Cj (confidence level Confij),
Condition 1. Confidence level is the same value Confif = Confii
Condition 2. One of the cases (i) to (vi).

  (I) If Ci = Cj and Pi is {Comij} and Pj is {Comij, pj1, pj2,..., Pjn}, Ri is deleted. (However, Comij is an item common to Pi and Pj.) That is, the difference between two association rules is only an increase in the condition part of one association rule.

For example,
Ri: {specific account-correspondence} ⇒ application certainty: 0.75
Rj: {specific account-corresponding, bad} ⇒ sign up Certainty: 0.75
In the case of Ri, Ri is deleted.

  (Ii) Ri is deleted when Pi = Pj and Ci is {Comij} and Cj is {Comij, cj1, cj2,..., Cjn}. (However, Comij is an item common to Ci and Cj.) That is, the difference between two association rules is only an increase in the conclusion part of one association rule.

For example,
(R2-1) {Open (not)} ⇒ {Apply (da)} Certainty: 1.00
(R3-2) {Open (not)} ⇒ {Specific account, apply (da)} Certainty factor: If 1.00, delete (R2-1).

(Iii) When Ci = Cj and Pi is {Comij, Pi1, Pi2} and Pj is {Comij, Pj1,..., Pjn},
If pi1 = word A and pi2 = word B and pj1 = word A-word B, Ri is deleted. (However, Comij is an item common to Pi and Pj.) That is, as a difference between the two association rules, the item in the condition part depends on a dependency relationship or a word co-occurrence relationship.

For example,
(R2-6): {specific account-apply (da)} ⇒ {open (no)} certainty: 0.67
(R3-1): {specific account, apply (da)} ⇒ {open (not)} certainty factor: 0.67, delete (R3-1).

(Iv) When Pi = Pj and Ci is {Comij, ci1, ci2} and Cj is {Comij, cj1,..., Cjn}
If ci1 = word A and ci2 = word B, cj1 = word A-word B, Ri is deleted. (However, Comij is an item common to Ci and Cj.) That is, as a difference between the two association rules, the item in the conclusion part depends on a dependency relationship or a word co-occurrence relationship.

For example,
(R2-5) {open (not)} ⇒ {specific account-apply (da)} certainty: 1.00
(R3-2) {Open (not)} ⇒ {Specific account, apply (da)} Certainty factor: If 1.00, delete (R3-2).

(V) Ci: Cj and Pi is {Comij, pi1}, Pj is {Comij, pj1,. . pjn},
If pi1 = word A and (pj1 = word A-word B or pj1 = word B-word A), Ri is deleted. (However, Comij is an item common to Pi and Pj.)
That is, the difference between the two association rules is whether the item in the condition part of the association rule is a word item or a dependency item including the word.

For example,
(R2-14) {Bad} ⇒ {Apply (da)} Confidence: 0.67
(R2-16) {Correspondence-Bad} ⇒ {Apply (da)} Confidence: 0.67
Then, (R2-14) is deleted. Similarly, (R2-12) is also deleted.

(Vi) When Pi = Pj and Ci is {Comij, ci1} and Cj is {comij, cj1,..., Cjn},
If ci1 = word A and (cj1 = word A-word B or cj1 = word B-word A), Ri is deleted. (However, Comij is an item common to Ci and Cj.)
That is, as a difference between the two association rules, whether the item of the conclusion part of the association rule is a word item or a dependency item including the word.

For example,
(R2-4) {open (not)} ⇒ {specific account} certainty: 1.00
(R2-5) {Open (not)] ⇒ {Specific account-Apply (da)} Confidence: If 1.00, delete (R2-4).

  As described above, when the association rule is extracted, the process proceeds to step 160 in FIG. 3 and the display unit 6 displays the association rule that has not been deleted in step 1150. As a result, as shown in FIG. 15, the next association rule with no strikethrough is output.

(A-3) Effects of the First Embodiment According to the first embodiment, unlike the case where only the dependency relationship is used as an item, by omitting the subject or object specific to the text data by considering the word. The corresponding association rule can be extracted.

  For example, in the case where only the dependency relationship is an item, the subject of “open (not)” of the data ID 2 is omitted, so the dependency relationship is not set. Since both of the dependency relationships are items, association rules such as {specific account-apply (da)} ⇒ {open (no)} can be extracted.

(B) Second Embodiment Next, a second embodiment of the information analysis system, information analysis method, and information analysis program of the present invention will be described with reference to the drawings.

  In the second embodiment, a method for outputting an association rule will be described.

(B-1) Configuration of Second Embodiment FIG. 16 is a block diagram showing an example of the internal configuration of a data analysis device 7B of the second embodiment.

  The configuration shown in FIG. 16 is different from the first embodiment in the functional configuration of the display unit 6. In the following, the functional configuration of the display unit 6 will be described in detail and described in the first embodiment. Detailed description of the configuration is omitted.

  FIG. 17 is a diagram showing a display image of the association rule displayed by the display unit 6. As shown in FIG. 17, each node corresponds to one item, and a branch between the nodes is a condition part of the association rule. And the relationship with the conclusion part. In FIG. 17, it is displayed as an undirected graph, and it is not shown which is a condition part and which is a conclusion part, but it may be a directed graph so that it becomes clear.

  For example, the branch between “correspondence-bad” and “specific account” in FIG. 17 is either association rule “{corresponding-bad} → {specific account}” or “{specific account} → {corresponding-bad}”. Indicates that exists.

  As illustrated in FIG. 16, the display unit 6 includes a graph element operation unit 601, a graph display unit 602, a graph node temporary storage unit 603, and a graph branch temporary storage unit 604.

  The graph element operation unit 601 takes in and executes operations necessary for displaying the association rule in the graph display shown in FIG.

  Here, the operations executed by the graph element operation unit 601 are as shown in the following (operation 1) to (operation 4). The graph element operation unit 601 deletes a branch or between nodes. A new branch is added or a node representing a superordinate concept is newly added.

  (Operation 1) Conceptually similar nodes are displayed nearby to support idea generation for the user. For example, the dependency item {correspondence-bad} and the word items “correspondence” and “bad” are displayed in the vicinity.

  (Operation 2) When displaying a graph, it is important whether there is a relationship between items, and the difference in reliability is not so important.

Therefore,
(R2-1l) {Correspondence} ⇒ {Apply (da)} Certainty: 0.75
(R2-16) {Correspondence-Bad} ⇒Apply (da)} Confidence: 0.67
If the item in the condition part of the association rule is a word item or a dependency item including the word, the association rule for the word item is not displayed as the difference between the two association rules. . In this example, (R2-1l) is not displayed.

  Further, the same applies to the case where the item in the conclusion part of the association rule is a word item or a dependency item including the word as a difference between the two association rules.

  (Operation 3) When displaying the dependency item “response-bad”, the association rules “{response] → {bad}” “{bad} → {response}” are not displayed.

  (Operation 4) Verbs, adjectives and adjective verbs that differ only in intention information such as “possible” and “denial” are also displayed nearby. For example, the items “accident-encounter”, “encounter (ta)”, and “accident-encounter (ta)” in FIG.

  In (Operation 1) and (Operation 4), as a method of displaying items close to each other, a branch is added between items related by these operations, a node representing a superordinate concept is newly added, Set a branch to.

  The graph display unit 602 creates a graph according to the result of the graph element operation unit 601.

(B-2) Operation of Second Embodiment Next, the operation of the second embodiment will be described with reference to the drawings.

  18 and 19 are flowcharts showing display processing by the display unit 6 according to the second embodiment. In the following, the association rule output by the display unit 6 is an association rule composed of an item set of size 1 shown in FIG.

  When an association rule is given to the display unit 6 by the association rule extraction unit 5, the graph element operation unit 601 of the display unit 6 uses each item of the condition part and the conclusion part of the association rule as a node of the graph, and displays each item as a graph. The data is stored in the node temporary storage unit 603 (step 2000).

  FIG. 20 shows an example of storage stored in the graph node temporary storage unit 603. As illustrated in FIG. 20, the graph node temporary storage unit 603 includes items of “node”, “number of appearances”, and “corresponding document ID”.

  The “node” item stores all items appearing in the condition part and the conclusion part of the association rule. The “appearance number” item and the “corresponding document ID” item are acquired from the item set temporary storage unit 505 of size 1 in FIG. 10 and stored.

  Further, the graph element operation unit 601 stores the relationship between the condition part and the conclusion part in the graph branch temporary storage unit 604 as a branch connecting the nodes of the graph (step 2000).

  FIG. 21 shows an example of storage stored in the graph branch temporary storage unit 604. As illustrated in FIG. 20, the graph branch temporary storage unit 604 includes items of “node 1”, “node 2”, “type”, “number of appearances”, and “corresponding document 1D”.

  The “node 1” and “node 2” items represent two nodes of the branch. The “type” item represents a branch type, and stores information on whether the branch is created from an association rule or whether it is created from (operation 1) or (operation 4). The “appearance number” item and the “corresponding document ID” item are acquired from the item set temporary storage unit 505 of size 2 in FIG. 11 and stored.

  The graph element operation unit 601 extracts a node stored in the graph node temporary storage unit 603, and determines whether the extracted node is a dependency item or a word item (step 2020).

  If the selected node is a dependency item, the process proceeds to step 2030. If the selected node is a word item, the process proceeds to step 2070.

  When the selected node is a dependency item, the graph element operation unit 601 determines whether the dependency source word or the dependency destination word of the node exists as a word item in the graph node temporary storage unit 603, and the graph When the node temporary storage unit 603 exists, a branch is set between two nodes having the dependency item and the source word or destination word as nodes (step 2030).

  FIG. 22 shows the stored contents of the graph branch temporary storage unit 604 updated by the processing of step 2030. Here, regarding the branch to be added, “node 1” is “word item”, “node 2” is “dependency item”, and the value of the “branch type” item is “dependency”. The “appearance number” item and the “corresponding document ID” item refer to the item set temporary storage unit 505 of size 1 in FIG. 10 and store the value of the word item.

  For example, taking the dependency item {correspondence-bad} in FIG. 22 as an example, the graph element operation unit 601 includes the dependency source word {correspondence} and the dependency destination word {bad} in the graph node temporary storage unit 603. If confirmed, the dependency source word {correspondence} is stored in "node 1", the dependency item {response-bad} is stored in "node 2", and "dependency" is stored in "branch type". In addition, items stored in the item set temporary storage unit 505 are stored in the “appearance count” and “corresponding document ID”.

  If it exists as a word item in the graph node temporary storage unit 603 and a branch can be set, the process proceeds to step 2050. If it does not exist as a word item in the graph node temporary storage unit 603, the process proceeds to step 2040.

  In step 2040, if the source word or destination word of the relevant node matches the source word or destination word of another dependency item in the graph node temporary storage unit 603, the matched word is added as a node. And set a branch between the two nodes.

  For example, if {correspondence-bad} and {correspondence-slow} exist in the graph node temporary storage unit 603, the word item “correspondence” is newly added to the graph node temporary storage unit 603, and “correspondence” and “ A branch is set between “response” and “response-slow” while “response-bad”.

  In Step 2050, (Operation 4) is performed. That is, if the source word or destination word of the node is a verb, an adjective or an adjective verb and includes intention information such as “possible” and “denial”, the word item not including the intention information is stored in the graph node temporary storage unit. Register to 603 and set a branch between the word item. However, if the word item is already registered in the graph node temporary storage unit 603, only the branch is stored in the graph branch temporary storage unit. Then, it returns to step 2010 and repeats processing.

  FIG. 23 shows an example of the stored contents of the graph node temporary storage unit 603 updated by the processing of step 2050.

  Here, an example of the added node is the word “apply” obtained by removing the intention information from the dependency word “apply (da)” of the dependency item “specific account—apply (da)” in the first line of FIG. 20. It is. In the “appearance number” item and “corresponding document ID” item, the “appearance number” item and “corresponding document ID” item of “specific account-apply (da)” in FIG. 20 are set as they are. However, when the word “apply” excluding the intention information is already registered, the appearance number item and the corresponding document ID item are added to the already registered values.

  FIG. 22 shows an example of the contents stored in the graph branch temporary storage unit 604 updated by the processing in step 2050.

  Here, in the example of the added branch, “node 1” is an item that does not include intention information, “node 2” is the node, and the value of the “branch type” item is “verb”. In the “appearance number” item and the “corresponding document ID” item, the appearance number item of the node and the corresponding document ID are set as they are.

  In Step 2060, (Operation 4) is performed. If the node is a verb, an adjective or an adjective verb and includes intention information such as “possible” or “denial”, a word item not including the intention information is registered in the graph node temporary storage unit, and Set a branch to. However, if the word item is already registered in the graph node temporary storage unit, only the branch is stored in the graph branch temporary storage unit. Return to step 2010.

  FIG. 24 shows an example of the contents stored in the graph node temporary storage unit 603 updated by the processing of step 2060.

  Here, the added node examples are the words “Apply” and “Open” obtained by removing the intention information from the word items “Apply (da)” and “Open (not)” in the second and third lines of FIG. . In the “appearance number” item and the “corresponding document ID” item of the word “open”, the appearance number item and the corresponding document ID item of “open (none)” in FIG. 20 are set as they are. Since the word “apply” is already registered, the “appearance number” item and the “corresponding document ID” item are added to the already registered values.

  The graph element operation unit 601 performs the processing from the next step 2080 on each element of the graph branch temporary storage unit 604, and when the processing is completed for all the elements of the graph branch temporary storage unit 604, the process proceeds to step 2120 ( Step 2070).

  When the graph element operation unit 601 selects the branch to be processed from the graph branch temporary storage unit 604, if the “type” item of the selected branch is “rule”, the process proceeds to step 2090; Returning to 2070, the processing is repeated (step 2080).

  When the “type” item of the selected branch is “rule”, the graph element operation unit 601 proceeds to step 2100 if both “node 1” and “node” set to the processing target branch are word items. Otherwise, return to Step 2110 and repeat the process (Step 2090).

  When both “node 1” and “node 2” of the branch to be processed are word items, the graph element operation unit 601 performs (operation 3). That is, the graph element operation unit 601 determines that the dependency relationship “node of the selected branch 1-node 2 of the selected branch” or “node 2 of the selected branch 2—node 1 of the selected branch” is the other branch. If it exists at the node, the branch to be processed is deleted (step 1100).

  For example, FIG. 22 shows an example of the contents stored in the graph branch temporary storage unit 604 updated by the processing in step 2100. In FIG. 22, the first and fourth lines are deleted. In the first line, “node 1” and “node 2” are both word items, and since there is a dependency relationship “correspondence-bad” on “node 1” in the seventh line, it is deleted. Similarly, the fourth line has been deleted because there is a dependency relationship “specific account—apply” at “node 2” on the second line.

  When “node 1” and “node 2” of the branch to be processed are not word items, the graph element operation unit 601 performs (operation 2). That is, the graph element operation unit 601 is different from the branch to be processed only in whether the item of “node 1” or “node 2” is a word item or a dependency item including the word. If there is a branch, delete the node for the word item. Thereafter, the process returns to step 2070 and the process is repeated.

  For example, FIG. 22 shows an example of the contents stored in the graph branch temporary storage unit 604 updated by the processing in step 2110. In FIG. 22, the third and sixth lines are deleted. Lines 3 and 5 differ only in node 1 "specific account" on line 3 and node 1 "specific account-apply for" on line 6; It is. Therefore, the third line in which node 1 is a word item is deleted.

  In the 5th and 8th lines, the node 1 “correspondence” in the 5th line is different from the node 1 “correspondence-bad” in the 8th line, and the node 2 is common to “apply”. . Therefore, the fifth line in which node 1 is a word item is deleted.

  When the graph element operation unit 601 has processed all branches, the graph display unit 602 displays the graph in a graph format based on the graph node temporary storage unit 603 and the graph branch temporary storage unit 604 (step 2120). .

  For example, a method described in Non-Patent Document 3 or the like can be used as a method of displaying the graph format by the graph display unit 602.

  As a result, a graph representation as shown in FIG. 25 is output. In FIG. 25, the solid line branch in the figure represents the relationship according to the association rule, and the broken line represents the other relationship. Compared to FIG. 17, branches representing association rules are reduced, and the positions of the word item and the dependency item including the word item are closer.

(B-3) Effect of Second Embodiment According to the second embodiment, when displaying association rules in a graph format, by displaying nodes that are semantically similar to each other nearby, the user can Helps you find important rules.

  In addition, the difference between the two association rules is small, and by displaying only the more limited association rules, the user can prevent overlooking other important association rules.

(C) Third Embodiment Next, a third embodiment of the information analysis system, the information analysis method, and the information analysis program of the present invention will be described with reference to the drawings.

  Since an item set composed only of words has higher support and certainty than dependency items, a large number of association rules composed only of word items are generated. If many such rules are output, the user is more likely to miss important rules. Therefore, in the third embodiment, association rules are limited so that a large number of association rules composed only of word items are not output.

  Specifically, for example, in association rules that include dependency items, such as {mother-apply (da)} ⇒ {specific account-apply (da)}, a low percentage of support and confidence are displayed. An association rule composed only of word items, such as {mother} ⇒ {advertise}, is displayed with a higher degree of support and certainty.

  Therefore, in the third embodiment, different support levels and certainty levels can be set for the dependency item and the word item. As a result, it is possible to prevent a large number of association rules composed only of word items from being displayed.

(C-1) Configuration and Operation of Third Embodiment Since the configuration of the third embodiment corresponds to the configuration of the first embodiment shown in FIG. 1, the third embodiment will be described below using FIG. A form is demonstrated.

  The third embodiment is different from the first embodiment in that, in the first embodiment, the input unit 1 takes in “minimum support” and “minimum certainty”, but in the third embodiment, , “Minimum support for word items only”, “minimum support for other cases”, “minimum confidence only for word items”, and “minimum confidence in other cases”.

  In addition, the candidate item set calculation unit 503 of the third embodiment selects an item set to continue processing based on “minimum support level only for word items” and “minimum support level in other cases”. It is.

  Furthermore, the rule creation unit 504 of the third embodiment is to select the obtained association rule based on “minimum certainty only for word items” and “minimum certainty in other cases”.

  Also, in FIG. 3, the third embodiment differs from the first embodiment in step 100 and step 150. Therefore, in the following, the characteristic processing of the third embodiment corresponding to these steps is performed. The description will be exchanged as 300 and 350.

  First, the input unit 1 inputs data as in the first embodiment of FIG. At this time, the input unit 1 includes “minimum support level only for word items”, “minimum support level for other cases”, “minimum confidence level for only word items”, “minimum confidence level for other cases”, The “maximum size of the item set to be created” is fetched and given to the association rule extraction unit 5 (step 300).

  For example, “minimum support for word items only” = 2, “minimum support for other cases” = 2, “minimum confidence only for word items” = 0.8, “minimum confidence in other cases” ] = 0.6, “maximum size of item set to be created” = 3.

  In step 350, the association rule extraction unit 5 performs an association rule creation process shown in FIG.

  Here, FIG. 26 shows processing in the association rule extraction unit 5 of the third embodiment, and processing corresponding to the processing in FIG. 3 is given the same reference numerals.

  First, generation of an item set of size 1 by the candidate item set generation unit 501 and calculation of the number of appearances of an item set of size 1 by the candidate item set calculation unit 503 are the same as the processing of the first embodiment.

  The candidate item set deletion unit 502 compares the word item in each item set with the “minimum support level of only word items” input to the input unit 1, and the number of occurrences of each word item is “minimum of only word items” If it is less than “support level”, the word item is deleted from the item set temporary storage unit 505 (step 3020).

  Further, the candidate item set deletion unit 502 compares the dependency item of each item set with the input “minimum support level in other cases”, and the number of appearance of each dependency item is “other than that” If it is less than the “minimum support level”, the dependency item is deleted from the item set temporary storage unit 505 (step 3020).

  After that, the control unit 500 sets the counter n to 2, the candidate item set generation unit 501 creates an item set of size 2, and the candidate item set deletion unit 502 sets the subset as in the first embodiment. Is deleted (steps 1030 to 1060).

  Then, the candidate item set calculation unit 503 calculates the support level for each item set of size 2, and among the item sets of size 2, for all items whose constituent elements are word items, the support level and “ Compared with “minimum support level of only word items”, an item set less than “minimum support level of only word items” is deleted from the item set temporary storage unit 505 (step 3070).

  In addition, the candidate item set calculation unit 503 has a support level and “minimum support level in other cases” for items having at least one dependency item among the size 2 item sets. And an item set less than “minimum support in other cases” is deleted from the item set temporary storage unit 505 (step 3070).

  When the counter n is equal to the “maximum size of the item set to be created” input from the input unit 1 or when all the item sets are less than the minimum support level, Proceed to 1100, otherwise proceed to step 1090, add 1 to the counter n, return to step 1050 and repeat the process (step 3080).

  As in the first embodiment, when all the item sets stored in the item set temporary storage unit 505 are divided into the condition part and the conclusion part, all of the condition part and the conclusion part of each item set are stored. The certainty factor for the combination is calculated (steps 1100 to 1130).

  Then, when the condition part and the conclusion part of each item set are all word items, the certainty of the combination is compared with the “minimum certainty only for word items” and the certainty is “minimum certainty only for word items”. If it is less than “degree”, the combination is deleted (step 3140). That is, it is not adopted as an association rule.

  In addition, when there is at least a dependency item in either the condition part or the conclusion part of each item set, the reliability of the combination is compared with the "other minimum confidence" and the confidence is If it is less than “minimum certainty”, the combination is deleted (step 3140).

  Thereafter, the process returns to Step 1120 or Step 1100, and when all combinations of all item sets are completed, duplicated association rules are selected and the process ends (Step 3150).

  Here, FIG. 27 shows an association rule made from a size 2 item set. In FIG. 27, the strikethrough is a rule that has been deleted compared to the first embodiment. Since the “minimum certainty only for word items” is set high (set to 0.8), the association rule composed of word items is deleted.

(C-2) Effects of Third Embodiment As described above, according to the third embodiment, the same effects as those of the first embodiment can be obtained.

  In addition, according to the third embodiment, the support item and the word item can be set to have different support and certainty, so that the support item is high and the word item is likely to appear a lot. Output of association rules can be suppressed. This can prevent overlooking other important association rules.

(D) Fourth Embodiment Next, a fourth embodiment of the information analysis system, information analysis method, and information analysis program of the present invention will be described with reference to the drawings.

  In the first embodiment, not only dependency items but also word items are set, so that an item set that is not very meaningful may be generated.

  For example, the item set {mother, advertise (is)} is an item set that combines the word items “mother” and “advertise (is)”.

  On the other hand, in the item set {bad, apply (da)} that combines the word items "bad" and "apply (da)", the guess is that something bad happened when you applied Working, not necessarily the wrong item set.

  Therefore, in the fourth embodiment, an item set having an erroneous meaning caused by combining word items in which one is a noun and the other is a predicate (verb, adjective or adjective verb) is prevented from being generated as much as possible. .

  For this reason, when creating an item set in which one is a noun and the other is only a word of predicate, it is assumed that at least one dependency relationship exists between the words. For example, to create an item set {mother, advertise (is)} that combines the word items “mother” and “advertise (is)”, the original input data contains “mother-advertise (is)”. Only when there is a dependency relationship.

(D-1) Configuration and Operation of Fourth Embodiment The configuration of the fourth embodiment corresponds to the configuration of the first embodiment shown in FIG. Further, the fourth embodiment differs from the first embodiment in the association rule extraction processing by the association rule extraction unit 5.

  Therefore, in the following, association rule extraction processing of the association rule extraction unit 5 of the fourth embodiment will be described with reference to the flowchart of FIG.

  In FIG. 28, the fourth embodiment is that step 4055 shown below is performed before the process of step 1060 of the first embodiment, and other processes are the same as those of the first embodiment. It is the same.

  In step 4055, the candidate item set deletion unit 502 performs the following check only when an item set of size 2 is being created.

  For each item set {A, B} of size 2, when elements A and B of the item set are words, one of A or B is a noun, and the other is a predicate, FIG. The candidate item set deletion unit 502 refers to the dependency relationship by the parsing unit 3 as shown, and if the relationship “AB” or “BA” is not satisfied as the dependency relationship, the candidate item set deletion unit 502 selects the item set {A , B}.

  For example, if the “minimum support” is 1, when creating an item set of size 2 from the data in FIG. 10, the item set {mother, advertise (is)} It is deleted from the item set temporary storage unit 505. This is because there is no dependency relationship of “mother-advertise” or “advertise-mother” in FIG.

  However, in this check, the intention information of the predicate is not asked. For example, if “mother-advertise (not)” and “mother-advertise (ta)” exist, the condition is satisfied.

  On the other hand, the item set {specific account, opened (none)} is not deleted from the size 2 item set temporary storage unit 505. This is because there is a dependency relationship “specific account-open (not)” on the eighth line in FIG. 6.

(D-2) Effect of Fourth Embodiment As described above, according to the fourth embodiment, the same effect as that of the first embodiment can be obtained.

  In addition, according to the fourth embodiment, by referring to the dependency relationship, a combination of word items in which one is a noun and the other is a predicate (a verb, an adjective or an adjective verb) is used. It is possible to prevent the generated item set from being generated as much as possible.

(E) Other Embodiments (1) In the first embodiment, intention information indicating denial or continuation is set in the predicates such as verbs and adjectives in FIG. 7 and FIG. Also good.

  (2) The concept hierarchy may be used when creating the item of the first embodiment. For example, when there is an upper / lower conceptual relationship between the “account” and the “specific account”, the data ID 3 in FIG. 8 may be added to the next item.

"Account-Apply (da)", "Account-Open (not)", "Account"
(3) In the first embodiment, the minimum support level and the minimum confidence level are input, but may be fixed values by the system. In addition to the minimum support level and the minimum certainty level, a list value or the like may be input. Alternatively, the upper limit of the association rule to be output may be defined by allowing the maximum support level and the maximum certainty level to be input.

  (4) The word items registered in step 140 of the first embodiment may be limited to words that do not appear in the dependency source word or the dependency destination word of the dependency item. Moreover, you may restrict | limit only to a verb, an adjective, or only a noun.

  (5) In creating the items of each data in FIGS. 7 and 8 of the first embodiment, attribute data of the data may be added. For example, as long as input data obtained from a call center, the customer's age and gender information may be used as an item.

  (6) Although the branch is an undirected graph in the second embodiment, it may be a directed graph from the condition part to the conclusion part.

  (7) In the second embodiment, among the data in the graph branch temporary storage unit added by the graph element operation unit 601, branches other than the type item “rule” are simply calculated as the attractive force between the nodes, and are displayed. You don't have to.

(8) In the second embodiment, a concept hierarchy may be used when a new branch is added by the graph element operation unit 601. For example, when there is an upper / lower conceptual relationship between “account” and “specific account”, FIG.
Node 1 item: "Account"
Node 2 item: “Specific Account”
Type: "Concept"
Number of occurrences: Number of occurrences of the word item “account” and its subordinate concept word items Applicable document ID: Data of a document in which the word item “account” and its subordinate concept word items appear may be added.

  (9) When calculating the minimum support level, a weighted support level using the number of appearances in the same data may be used.

  (10) In the first to fourth embodiments described above, the system according to the present invention has been described as being realized by the same device, but can be realized by distributed processing by a plurality of separate devices that can be connected to each other. It is good as a thing.

  (11) The system described in the first to fourth embodiments described above is a function realized by an information processing apparatus such as a personal computer or a workstation, but its substance is a process that can be executed by the information processing apparatus. It is a program. The processing program of the system described in the first to fourth embodiments may be stored in a computer-readable storage medium, stored in a hard disk, or transmitted through a network, for example. .

  (12) The arrangement of components of the system according to the present invention and the order of processing flows are not limited to those described in the first to fourth embodiments.

It is a block diagram which shows the internal structure of the data analyzer of 1st Embodiment. It is explanatory drawing which shows the example of a dependency relationship. It is a flowchart which shows operation | movement of the data analyzer of 1st Embodiment. It is a figure which shows the example of input data of 1st Embodiment. It is a figure which shows the result of the morphological analysis process of 1st Embodiment. It is a figure which shows the result of the syntax analysis process of 1st Embodiment. It is a figure which shows the item creation result by the dependency relation of 1st Embodiment. It is a figure which shows the item creation result by the morpheme of 1st Embodiment. It is a flowchart which shows the association rule extraction process of 1st Embodiment. It is a figure which shows the example of a storage content of the item set temporary storage part of 1st Embodiment. It is a figure which shows the example of a storage content of the item set temporary storage part of 1st Embodiment. It is a figure which shows the association rule made from the item set of 1st Embodiment. It is a figure which shows the example of a storage content of the item set temporary storage part of 1st Embodiment. It is a figure which shows the association rule made from the item set of 1st Embodiment. It is a figure which shows the association rule extracted of 1st Embodiment. It is a block diagram which shows the internal structure of the data analyzer of 2nd Embodiment. It is a figure which shows the example of a graph display of the association rule of 2nd Embodiment. It is a flowchart which shows operation | movement of the display part 6 of 2nd Embodiment. It is a flowchart which shows operation | movement of the display part 6 of 2nd Embodiment. It is a figure which shows the example of the storage content of the graph node temporary storage part of 2nd Embodiment. It is a figure which shows the example of the storage content of the graph branch temporary storage part of 2nd Embodiment. It is a figure which shows the example of the storage content updated of the graph branch temporary storage part of 2nd Embodiment. It is a figure which shows the example of the storage content updated of the graph node temporary storage part of 2nd Embodiment. It is a figure which shows the example of the storage content updated of the graph branch temporary storage part of 2nd Embodiment. It is a figure which shows the example of a graph display of the association rule of 2nd Embodiment. It is a flowchart which shows the association rule extraction process of 3rd Embodiment. It is a figure which shows the association rule made from the item set of 3rd Embodiment. It is a flowchart which shows the association rule extraction process of 4th Embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Input part, 2 ... Morphological analysis part, 3 ... Syntax analysis part, 4 ... Item generation part, 5 ... Association rule extraction part, 500 ... Control part, 501 ... Candidate item set generation part, 502 ... Candidate item set deletion part 503 ... Candidate item set calculation unit 504 ... Rule creation unit 505 ... Item set temporary storage unit 6 ... Display unit 601 ... Graph element operation unit 602 ... Graph display unit 603 ... Graph node temporary storage unit 604 ... Graph branch temporary storage unit, 7A and 7B ... Data analysis device.

Claims (9)

In an information analysis system that creates a correlation rule based on each component of a plurality of input text information and outputs a useful correlation rule,
Morphological analysis means for performing morphological analysis on each text information,
A syntax analysis means for performing syntax analysis on each of the above text information;
Item creation means for creating a morphological analysis result and / or a syntax analysis result of each text information as an item to be analyzed by the correlation rule;
Item set creation means for creating one or more item sets using the one or more items created by the item creation means;
Item set deletion means for checking the item sets created by the item set creation means and deleting the item set having items having semantically inclusive relations as elements,
Item set calculation means for calculating the co-occurrence appearance frequency for each item set,
Correlation rule creating means for creating one or more correlation rules based on the co-occurrence frequency of each item set calculated by the item set calculating means;
An information analysis system comprising: display means for displaying each correlation rule created by the correlation rule creation means.   The information analysis system according to claim 1, wherein the item set deletion unit deletes a word item including the same character string as the dependency source character string or the dependency destination character string of the dependency item.   The correlation rule creating means divides each created correlation rule into a condition part and a conclusion part, a relationship between the divided condition part and the conclusion part, and the divided condition part and the conclusion. The information analysis system according to claim 1 or 2, wherein an association rule with few restrictions is deleted based on a certainty factor when co-occurring with a part.   The display means displays each item as a node, and displays the relationship between the condition part of the correlation rule and the conclusion part in the form of a graph. The nodes having similar concepts between the items are displayed. The information analysis system according to any one of claims 1 to 3, wherein the information is displayed close to the distance.   5. The correlation rule creating means selects the useful correlation rule using usefulness judgment information for judging the usefulness of each created correlation rule. Information analysis system described.   6. The information analysis system according to claim 5, wherein the usefulness determination information is defined according to a type of the item.   The information analysis system according to claim 1, wherein the item set creation unit creates an item set having a dependency relationship between word items constituting the item set. In an information analysis method for creating a correlation rule based on each component of a plurality of input text information and outputting a useful correlation rule,
A morpheme analysis step in which the morpheme analysis means performs morpheme analysis on each text information;
A parsing step in which the parsing means performs parsing on each of the text information;
An item creating step for creating a morphological analysis result and / or a syntax analysis result of each text information as an item to be analyzed by the correlation rule;
An item set creation step in which the item set creation means creates one or more item sets using the one or more items created by the item creation means;
An item set deletion step in which an item set deletion unit deletes the item set having an item having a semantically inclusive relationship as an element by comparing each item set created by the item set creation unit,
Item set calculation means, the item set calculation step for calculating the co-occurrence appearance frequency for each item set,
A correlation rule creating step for creating one or more correlation rules based on the co-occurrence appearance frequency of each item set calculated by the item set calculating unit;
A display means comprising: a display step of displaying each correlation rule created by the correlation rule creating means. In an information analysis program that creates a correlation rule based on each component of a plurality of input text information and outputs a useful correlation rule,
On the computer,
Morphological analysis means for performing morphological analysis on each text information,
Syntax analysis means for performing syntax analysis on each of the above text information;
Item creation means for creating a morphological analysis result and / or a syntax analysis result of each text information as an item to be analyzed by the correlation rule,
Item set creation means for creating one or more item sets using the one or more items created by the item creation means,
Item set deletion means for deleting each item set created by the item set creation means and deleting the item set having an item having a semantically inclusive relationship as an element,
Item set calculation means for calculating the co-occurrence appearance frequency for each item set,
Correlation rule creating means for creating one or a plurality of correlation rules based on the co-occurrence appearance frequency of each item set calculated by the item set calculating means;
An information analysis program that functions as display means for displaying each of the correlation rules created by the correlation rule creation means.

Images