Information processing device, information processing method, and program

Abstract

In order to more accurately generate a desired taxonomy, an information processing device (1) comprises: an acquisition unit (11) that acquires a sentence set and auxiliary information which pertains to the taxonomy to be constructed; a generation unit (12) that refers to the sentence set and generate a phrase; and a construction unit (13) that infers a relationship between a plurality of phrases including the phrase so as to construct the taxonomy. The auxiliary information is used by the generation unit (12) and / or the construction unit (13).

Description

[Technical field]

[0001] The present invention relates to a technology for generating a taxonomy. [Background technology]

[0002] There is a known technology for generating a taxonomy. A taxonomy is information that represents the relationships between multiple phrases, and is expressed as a hierarchical directed graph.

[0003] For example, Non-Patent Document 1 describes a technology for constructing a taxonomy with probabilities. This technology extracts phrases from a set of sentences, assigns probability values ​​to the relationships between the extracted phrases, and converts the graph between the phrases constructed based on the probability values ​​into a taxonomy.

[0004] For example, Non-Patent Document 2 describes a technology for generating a taxonomy by adding phrases extracted from a set of sentences to a seed taxonomy. A seed taxonomy is an incomplete, relatively small taxonomy that is generated manually or the like. [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] Wentao Wu et al., "Probase: A Probabilistic Taxonomy for Text Understanding", SIGMOD 2012 [Non-Patent Document 2] Jiaming Shen et al., "HiExpan: Task-Guided Taxonomy Construction by Hierarchical Tree Expansion", KDD 2018. Summary of the Invention [Problem to be solved by the invention]

[0006] However, the techniques described in Non-Patent Documents 1 and 2 may extract phrases from a sentence set that are not in line with the purpose of the taxonomy. In addition, these techniques may estimate relationships between phrases that are not in line with the purpose of the taxonomy. For this reason, these techniques have room for improvement in terms of accurately generating a desired taxonomy that meets the purpose.

[0007] One aspect of the present invention has been made in consideration of the above problems, and one example of a purpose of the present invention is to provide a technique for generating a desired taxonomy with greater accuracy. [Means for solving the problem]

[0008] An information processing device according to one aspect of the present invention includes an acquisition means for acquiring auxiliary information regarding a set of sentences and a taxonomy to be constructed, a generation means for generating a phrase by referring to the set of sentences, and a construction means for constructing the taxonomy by estimating relationships between a plurality of phrases including the phrase, and the auxiliary information is used in one or both of the generation means and the construction means.

[0009] An information processing method according to one aspect of the present invention includes obtaining auxiliary information regarding a set of sentences and a taxonomy to be constructed, generating phrases by referring to the set of sentences, and constructing the taxonomy by estimating relationships between a plurality of phrases including the phrases, and using the auxiliary information in one or both of generating the phrases and constructing the taxonomy.

[0010] A program according to one aspect of the present invention is a program for causing a computer to function as an information processing device, causing the computer to function as an acquisition means for acquiring auxiliary information regarding a set of sentences and a taxonomy to be constructed, a generation means for generating a phrase by referring to the set of sentences, and a construction means for constructing the taxonomy by estimating relationships between multiple phrases including the phrase, and using the auxiliary information in one or both of the generation means and the construction means. Effect of the Invention

[0011] According to one aspect of the present invention, a desired taxonomy can be generated with greater accuracy. [Brief description of the drawings]

[0012] [Figure 1] 1 is a block diagram showing a configuration of an information processing device according to a first exemplary embodiment. [Diagram 2] 1 is a flow chart showing the flow of an information processing method according to the first exemplary embodiment. [Diagram 3] FIG. 11 is a block diagram showing a configuration of an information processing device according to a second exemplary embodiment. [Figure 4] 11 is a flow chart showing the flow of an information processing method according to the second exemplary embodiment. FIG. [Diagram 5] FIG. 11 is a flow chart showing an example of a detailed flow of a process for generating a phrase in the second exemplary embodiment. [Figure 6] FIG. 11 is a flow diagram showing an example of a detailed flow of a process for estimating a relationship in the second exemplary embodiment. [Figure 7] FIG. 11 is a block diagram showing a configuration of an information processing device according to an exemplary embodiment 3. [Figure 8] FIG. 13 is a diagram illustrating an example of a taxonomy obtained in exemplary embodiment 3. [Figure 9] FIG. 11 is a flow chart showing the flow of an information processing method according to the third exemplary embodiment. [Figure 10] FIG. 11 is a flow chart showing an example of a detailed flow of a process for generating a phrase in exemplary embodiment 3. [Figure 11] FIG. 11 is a flow diagram showing an example of a detailed flow of a process for estimating a relationship in the second exemplary embodiment. [Figure 12] 1 is a block diagram showing an example of a hardware configuration of an information processing device according to each exemplary embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Example Embodiment 1 A first exemplary embodiment of the present invention will be described in detail with reference to the drawings. This exemplary embodiment is a basic form of the exemplary embodiments described later.

[0014] <Configuration of information processing device 1> The configuration of an information processing device 1 according to this exemplary embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the functional configuration of the information processing device 1.

[0015] As shown in Fig. 1, the information processing device 1 includes an acquisition unit 11, a generation unit 12, and a construction unit 13. The acquisition unit 11 is an example of a configuration for realizing an acquisition means recited in the claims. The generation unit 12 is an example of a configuration for realizing a generation means recited in the claims. The construction unit 13 is an example of a configuration for realizing a construction means recited in the claims.

[0016] The acquisition unit 11 acquires a set of sentences and auxiliary information related to the taxonomy to be constructed. The generation unit 12 generates phrases by referring to the set of sentences. The construction unit 13 constructs the taxonomy by estimating relationships between multiple phrases including the phrases generated by the generation unit 12. One or both of the generation unit 12 and the construction unit 13 use the auxiliary information.

[0017] (phrase) A phrase is represented by one or more words and represents a concept. For example, the phrase "pasta" represents the concept of a dough made mainly from wheat flour, which is one of the staple foods in Italian cuisine. The phrase "Italian cuisine" represents the concept of a cooking method that originated in Italy. In the following, the "concept represented by a phrase" will be simply referred to as a "phrase." Therefore, "there is a relationship between the concepts represented by two phrases" will be described as "there is a relationship between two phrases." For example, the phrase "pasta" is included in the phrase "Italian cuisine." Therefore, there is a parent-child relationship between these phrases.

[0018] (Taxonomy) A taxonomy is information that represents the relationship between multiple phrases. For example, a taxonomy is represented by a directed graph having a hierarchy. Specifically, a taxonomy includes information that represents a node that represents each phrase and an edge that connects two nodes having a parent-child relationship. The taxonomy to be constructed is a taxonomy that is output by the information processing device 1.

[0019] The hierarchy in the taxonomy defines the parent-child relationship between related phrases. For example, the parent-child relationship may represent a relationship in which the parent is a higher concept and the child is a lower concept. For example, the parent-child relationship may represent a relationship in which the parent is an organization and the child is a member. For example, the parent-child relationship may represent a geographical inclusion relationship.

[0020] In addition, parent-child relationships included in the same taxonomy may differ depending on the hierarchical level. For example, in a taxonomy related to soccer teams, the parent-child relationship between the first and second hierarchical levels may represent a geographical inclusion relationship (parent is a country or region, and child is a soccer team), while the parent-child relationship between the second and third hierarchical levels may represent a relationship between affiliated organizations and members (parent is a soccer team, and child is a player).

[0021] (Sentence set) A sentence set is a set of sentences that is referenced when generating phrases to be included in the taxonomy to be built.

[0022] (Supporting information) The auxiliary information is information about the taxonomy to be constructed. The auxiliary information may include a natural language sentence or a set of words.

[0023] For example, the auxiliary information may include information indicating the purpose of the taxonomy to be constructed. The purpose of the taxonomy may be, for example, the intention of generating the taxonomy or its use. The auxiliary information may also include information limiting phrases to be included in the taxonomy to be constructed. The auxiliary information may also include information indicating relationships between phrases in the taxonomy to be constructed.

[0024] An example of auxiliary information that includes information that limits a phrase is information that includes natural language sentences such as "taxonomy of ingredients that appear in the recipe" or "taxonomy of cooking actions that appear in the recipe."

[0025] In addition, examples of auxiliary information including information indicating relationships include natural language sentences such as "a taxonomy related to soccer teams, divided into countries and regions" and "a taxonomy related to soccer teams, including the players currently belonging to the teams."

[0026] <Flow of information processing method S1> The flow of the information processing method S1 executed by the information processing device 1 configured as above will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of the information processing method S1. As shown in Fig. 2, the information processing method S1 includes steps S11 to S13.

[0027] (Step S11) In step S11, the acquisition unit 11 acquires the sentence set and the auxiliary information. For example, the acquisition unit 11 may acquire one or both of the sentence set and the auxiliary information from a memory (not shown) of the information processing device 1, from another device connected via a network, or via an input device.

[0028] (Step S12) In step S12, the generation unit 12 generates a phrase by referring to the sentence collection. Note that "generating a phrase by referring to the sentence collection" may mean "extracting a phrase included in the sentence collection" or "predicting a phrase that is not necessarily included in the sentence collection based on the sentence collection." In this step, the generation unit 12 may or may not use auxiliary information to generate a phrase by referring to the sentence collection. However, at least one of step S12 and step S13 described later uses auxiliary information.

[0029] (When using auxiliary information) For example, the generation unit 12 generates a phrase similar to the auxiliary information by referring to a set of sentences. Specifically, the generation unit 12 calculates the similarity between the generated phrase and the auxiliary information by referring to the set of sentences. Furthermore, if the calculated similarity satisfies a predetermined condition, the generation unit 12 adopts the phrase as a phrase to be included in the taxonomy to be constructed, and if it does not satisfy the condition, does not adopt the phrase.

[0030] (When auxiliary information is not used) For example, the generation unit 12 generates phrases by referring to a collection of sentences, and sets the generated phrases as phrases to be included in the taxonomy to be constructed.

[0031] (Step S13) In step S13, the construction unit 13 constructs a taxonomy by estimating relationships between multiple phrases including the phrase generated in step S12. In this step, the construction unit 13 may or may not use auxiliary information to estimate the relationships between the multiple phrases. However, as described above, at least one of the above-mentioned step S12 and step S13 uses auxiliary information.

[0032] (When using auxiliary information) For example, the construction unit 13 estimates a relationship between multiple phrases that is consistent with the auxiliary information. Specifically, the construction unit 13 calculates information indicating the relationship between multiple phrases, and modifies the calculated information using the auxiliary information. The construction unit 13 also constructs a taxonomy by associating multiple phrases based on the modified information.

[0033] (When auxiliary information is not used) For example, the construction unit 13 calculates information indicating the relationship between a plurality of phrases, and constructs a taxonomy by associating the plurality of phrases based on the calculated information.

[0034] <Advantages of this exemplary embodiment> As described above, the information processing device 1 according to this exemplary embodiment uses auxiliary information in one or both of (i) a generation process for generating phrases by referring to a set of sentences, and (ii) a process for constructing a taxonomy by estimating relationships between multiple phrases including the generated phrase. As a result, when auxiliary information is used in the generation process, phrases included in the constructed taxonomy are more likely to conform to the content of the auxiliary information. Also, when auxiliary information is used in the construction process, the relationships between multiple phrases in the constructed taxonomy are more likely to conform to the content of the auxiliary information. Since the auxiliary information is information related to the taxonomy to be constructed, this exemplary embodiment can construct a desired taxonomy with greater accuracy.

[0035] Exemplary embodiment 2 A second exemplary embodiment of the present invention will be described in detail with reference to the drawings. Note that components having the same functions as those described in the first exemplary embodiment are given the same reference numerals, and the description thereof will be omitted as appropriate.

[0036] <Configuration of information processing device 2> The configuration of the information processing device 2 according to this exemplary embodiment will be described with reference to FIG. 3. FIG. 3 is a block diagram showing the functional configuration of the information processing device 2. As shown in FIG. 3, the information processing device 2 is a device that receives a sentence set D1 and auxiliary information F1 as input and outputs a taxonomy T1. The information processing device 2 includes an acquisition unit 21, a generation unit 22, and a construction unit 23. The acquisition unit 21 is configured similarly to the acquisition unit 11 in the exemplary embodiment 1. The generation unit 22 generates a plurality of phrases similar to the auxiliary information F1 with reference to the sentence set D1. The construction unit 23 estimates a relationship that is consistent with the auxiliary information F1 as a relationship between a plurality of phrases.

[0037] <Flow of information processing method S2> The flow of the information processing method S2 executed by the information processing device 2 configured as above will be described with reference to Fig. 4. Fig. 4 is a flow diagram showing the flow of the information processing method S2. As shown in Fig. 4, the information processing method S2 includes steps S21 to S23.

[0038] (Step S21) In step S21, the acquisition unit 21 acquires the sentence set D1 and the auxiliary information F1. Details of this step are as described in step S11 in the first exemplary embodiment.

[0039] (Step S22) In step S22, the generation unit 22 refers to the sentence set D1 and generates phrases similar to the auxiliary information F1. Details of this step will be described later.

[0040] (Step S23) In step S23, the constructor 23 estimates relationships between multiple phrases to be included in the taxonomy T1 that are consistent with the auxiliary information F1. The constructor 23 also constructs the taxonomy T1 by referring to the estimation result. Details of this step will be described later.

[0041] <Details of step S22> Details of step S22 will be described with reference to Fig. 5. Fig. 5 is a flow diagram showing an example of a detailed flow of the process of step S22. As shown in Fig. 5, the process of step S22 includes steps S221 to S222.

[0042] (Step S221) In step S221, the generation unit 22 generates two or more candidate phrases by referring to the sentence set D1. For the process of generating the candidate phrases, for example, the technique described in Non-Patent Document 1 can be used. However, the technique used for the process of generating the candidate phrases is not limited to this.

[0043] (Specific example of step S211) For example, as shown in Fig. 5, an example will be described in which the sentence set D1 includes the sentence "Here are some recipes using dairy products such as cheese. Chop the basil with a knife." In this case, the generation unit 22 generates candidate phrases "cheese," "dairy products," "recipe," "basil," and "knife" from the sentence set D1.

[0044] (Step S222) In step S222, the generation unit 22 extracts a plurality of phrases whose similarity to the auxiliary information F1 satisfies a predetermined condition from among two or more candidate phrases, and sets the plurality of phrases to be included in the taxonomy T1. In other words, the generation unit 22 narrows down the candidate phrases under the condition that they are similar to the auxiliary information F1. An example of the predetermined condition is a condition that the similarity between the candidate phrase and the auxiliary information F1 is equal to or greater than a threshold. The similarity can be calculated, for example, using a distributed representation. Specifically, the similarity may be a cosine similarity or Euclidean distance between the distributed representation of the candidate phrase and the distributed representation of the auxiliary information F1. However, the predetermined condition is not limited to this. Furthermore, the method of calculating the similarity is not limited to this. Hereinafter, narrowing down the candidate phrases under the condition that they are similar to the auxiliary information F1 is also simply referred to as "narrowing down using the auxiliary information F1".

[0045] (Specific example of step S222) For example, as shown in FIG. 5, an example will be described in which the auxiliary information F1 includes a natural language sentence "ingredients used in cooking." In this case, the generation unit 22 calculates a similarity of 0.9 between the auxiliary information F1 and "cheese" among the candidate phrases. Similarly, the generation unit 22 calculates similarities of 0.9, 0.1, 0.9, and 0.1 between the auxiliary information F1 and each of "dairy products," "recipe," "basil," and "knife." In this way, phrases that conform to the auxiliary information F1 "ingredients used in cooking" have a higher similarity to the auxiliary information F1 than phrases that do not.

[0046] Here, for example, assume that the threshold is 0.7. In this case, the generation unit 22 extracts the phrases “cheese,” “dairy products,” and “basil,” whose similarities with the auxiliary information F1 are equal to or greater than the threshold, as multiple phrases to be included in the taxonomy T1.

[0047] In this way, by narrowing down the search using the auxiliary information F1, the phrases "cheese," "dairy products," and "basil," which are similar to the auxiliary information F1, are adopted as phrases to be included in the taxonomy T1, while "recipe" and "knife," which are not similar to the auxiliary information F1, are not adopted.

[0048] 5, an example has been described in which candidate phrases are generated in step S221, and then the candidate phrases are narrowed down using the auxiliary information F1 in step S222. However, the present invention is not limited to this example, and the generation unit 22 may generate phrases similar to the auxiliary information F1 by referring to the auxiliary information F1 during the process of generating phrases from the sentence set D1.

[0049] <Details of step S23> Details of step S23 will be described with reference to Fig. 6. Fig. 6 is a flow diagram showing an example of a detailed flow of the process of step S23. As shown in Fig. 6, the process of step S23 includes steps S231 to S233.

[0050] (Step S231) In step S231, the construction unit 23 calculates a score indicating a relationship between a plurality of phrases included in the taxonomy T1. The score is a degree of certainty that a relationship exists, and is an example of "information indicating a relationship" described in the claims. For example, the technology described in Non-Patent Document 1 can be used for the process of calculating the score indicating the relationship. However, the technology used for the process of calculating the information indicating the relationship is not limited to this.

[0051] (Specific example of step S231) For example, a specific example of step S231 in the case where a plurality of phrases "cheese", "dairy products", and "basil" are generated in step S22 will be described. As an example, the construction unit 23 calculates 0.9 as a score indicating the relationship between the phrases "cheese" and "basil". This score is considered to indicate the relationship that the phrases "cheese" and "basil" are ingredients that go well together. Similarly, the construction unit 23 calculates 0.9 as a score indicating the relationship between the phrases "cheese" and "dairy products". This score is considered to indicate the relationship that "cheese" is included in "dairy products". Furthermore, the construction unit 23 calculates 0.1 as a score indicating the relationship between the phrases "basil" and "dairy products". This score indicates the possibility that these phrases are not related to each other.

[0052] (Step S232) In step S232, the constructing unit 23 modifies the score indicating the relationship between a plurality of phrases by using the auxiliary information F1. For example, the constructing unit 23 modifies the score by using the following formula (1).

[0053]

number

[0054] (Specific example of step S232) For example, the construction unit 23 obtains a similarity of 0.1 between the difference between the embedded representation of the phrase "cheese" and the embedded representation of the phrase "basil" and the embedded representation of the auxiliary information F1. This indicates that the similarity between the "relationship between ingredients that go well together" between the phrases "cheese" and "basil" and the auxiliary information F1 "ingredients used in cooking" is low. Then, the construction unit 23 adds the similarity of 0.1 to the score of 0.8 between the phrases "cheese" and "basil" according to formula (1) to correct the score to 0.9. Note that λ=1 is applied here.

[0055] Similarly, the construction unit 23 obtains a similarity of 0.9 between the difference between the embedded representation of the phrase "cheese" and the embedded representation of the phrase "dairy products" and the embedded representation of the auxiliary information F1. This indicates that the phrase "cheese" is highly related to being included in "dairy products" and has a high similarity with the auxiliary information F1 "ingredients used in cooking". The construction unit 23 then adds the similarity of 0.9 to the score of 0.9 between the phrases "cheese" and "dairy products" to revise the score to 1.8 according to formula (1).

[0056] Similarly, the construction unit 23 obtains a similarity of 0.1 between the difference between the embedded representation of the phrase "basil" and the embedded representation of the phrase "dairy products" and the embedded representation of the auxiliary information F1. This indicates that the relationship between the phrases "basil" and "dairy products" and the similarity with the auxiliary information F1 "ingredients used in cooking" is low. Then, the construction unit 23 adds the similarity of 0.1 to the score of 0.1 between the phrases "basil" and "dairy products" to correct the score to 0.2 according to formula (1).

[0057] In this way, if two phrases have a "relationship of ingredients that go well together" that is inconsistent with the auxiliary information F1, the score indicating such a relationship is modified to a relatively low value using the auxiliary information F1.

[0058] (Step S233) In step S233, the constructor 23 constructs the taxonomy T1 based on the corrected scores. As an example, the constructor 23 associates two phrases whose corrected scores are equal to or greater than a threshold.

[0059] (Specific example of step S233) For example, the construction unit 23 associates the phrases "cheese" and "dairy products" whose post-correction scores are equal to or greater than a threshold value of 1.0, and constructs a taxonomy T1 including these phrases. As shown in FIG. 6, the taxonomy T1 is represented by a directed graph including nodes N1 to N7 and edges E1 to E6. The nodes N1 to N7 correspond to the multiple phrases generated in step S22, respectively. For example, the node N3 corresponds to the phrase "dairy products", the node N5 corresponds to the phrase "basil", and the node N6 corresponds to the phrase "cheese". The edges E1 to E6 represent the existence of a "relationship of superordinate and subordinate concepts" between the two phrases connected to both ends. In FIG. 6, the edges E1 to E6 are represented by arrows, and the node connected to the root is the parent, and the node connected to the arrowhead is the child. For example, the edge E5 represents that the phrase "dairy products" corresponding to the node N2 connected to the root is the parent (superordinate concept), and the phrase "cheese" corresponding to the node N5 connected to the arrowhead is the child (subordinate concept).

[0060] In this way, the multiple phrases "cheese," "dairy products," and "basil" are related by the "relationship between superordinate and subordinate concepts of ingredients" which is similar to the auxiliary information F1, but are not related by the "relationship of ingredients that go well together" which is not similar to the auxiliary information F1.

[0061] <Advantages of this exemplary embodiment> As described above, the information processing device 2 according to this exemplary embodiment generates a plurality of phrases similar to the sentence set D1 as a plurality of phrases to be included in the taxonomy T1. As a result, the plurality of phrases included in the constructed taxonomy T1 are more likely to be consistent with the auxiliary information F1. In addition, the information processing device 2 according to this exemplary embodiment estimates a relationship that is consistent with the auxiliary information F1 as a relationship between a plurality of phrases, and constructs the taxonomy T1 based on the estimation result. As a result, the relationship included in the constructed taxonomy T1 is more likely to be consistent with the auxiliary information F1. Since the auxiliary information F1 is information related to the taxonomy T1 to be constructed, the present exemplary embodiment can construct the desired taxonomy T1 with greater accuracy.

[0062] Other Aspects of the Exemplary Embodiment In this exemplary embodiment, the auxiliary information F1 may not be used in either step S22 or S23. For example, the generation unit 22 may generate phrases from the sentence set D1 without using the auxiliary information F1 in step S22. Specifically, the generation unit 22 may omit the process of step S222 among steps S221 to S222 shown in Fig. 5. In this case, the generation unit 22 sets all the candidate phrases obtained in step S221 as a plurality of phrases to be included in the taxonomy T1.

[0063] Also, for example, the construction unit 23 may construct the taxonomy T1 without using the auxiliary information F1 in step S23. Specifically, the construction unit 23 may omit the process of step S232 among steps S231 to S233 shown in Fig. 6. However, the process of step S232 can be omitted only when the auxiliary information F1 is used in step S22. When the auxiliary information F1 is not used in step S22, the construction unit 23 executes the process of step S232 without omitting it.

[0064] In this way, even if auxiliary information F1 is used in one of steps S22 and S23 but not in the other, the desired taxonomy T1 can be constructed with greater accuracy than if auxiliary information F1 is not used in both steps.

[0065] Exemplary embodiment 3 A third exemplary embodiment of the present invention will be described in detail with reference to the drawings. Note that components having the same functions as those described in the first and second exemplary embodiments are denoted by the same reference numerals, and the description thereof will not be repeated.

[0066] <Configuration of information processing device 3> The configuration of the information processing device 3 according to this exemplary embodiment will be described with reference to FIG. 7. FIG. 7 is a block diagram showing the functional configuration of the information processing device 3. As shown in FIG. 7, the information processing device 3 is a device that receives a taxonomy T2, a sentence set D2, and auxiliary information F2 as input, and outputs a taxonomy T3. The information processing device 3 includes an acquisition unit 31, a generation unit 32, and a construction unit 33. The acquisition unit 31 is configured similarly to the acquisition unit 11 in the exemplary embodiment 1, and further acquires a taxonomy T2. The generation unit 32 refers to the sentence set D2 and generates a new phrase different from the phrase included in the taxonomy T2. The construction unit 33 estimates the relationship between the phrase included in the taxonomy T2 and the new phrase, thereby expanding the taxonomy T2 and constructing the taxonomy T3.

[0067] (Taxonomies T2, T3) Taxonomy T3 is a taxonomy output by information processing device 3, and is an example of a "taxonomy to be constructed" as described in the claims. Taxonomy T2 is a taxonomy that constitutes a part of taxonomy T3, and is an example of a "partial taxonomy" as described in the claims. Taxonomy T3 is constructed by extending taxonomy T2.

[0068] Taxonomy T2 includes phrases related to a specific domain. Hereinafter, including phrases related to a specific domain will also be referred to simply as "related to a specific domain."

[0069] FIG. 8 is a diagram illustrating an example of taxonomy T2. In this example, taxonomy T2 relates to the domain "news". Taxonomy T2 is represented by a directed graph including nodes N1 to N7 and edges E1 to E6. Nodes N1 to N7 correspond to a plurality of phrases in the domain "news", respectively. For example, node N2 corresponds to the phrase "weather", node N3 corresponds to the phrase "sports", and node N6 corresponds to the phrase "baseball". Note that edges E1 to E6 have been described using FIG. 6, and therefore detailed description will not be repeated.

[0070] (New phrase) The new phrase is a phrase that is different from any of the multiple phrases included in taxonomy T2 and is a phrase to be added to taxonomy T2. In other words, the new phrase is a phrase that is not included in taxonomy T2.

[0071] (Existing phrase) Hereafter, each phrase included in taxonomy T2 will be referred to as an "existing phrase" to distinguish it from new phrases.

[0072] (Sentence set D2) The sentence set D2 is referred to when generating phrases to be added to the taxonomy T2. The sentence set D2 is preferably a set of sentences related to the same domain as the existing phrases.

[0073] <Flow of information processing method S3> The flow of the information processing method S3 executed by the information processing device 3 configured as above will be described with reference to Fig. 9. Fig. 9 is a flow diagram showing the flow of the information processing method S3. As shown in Fig. 9, the information processing method S3 includes steps S31 to S33.

[0074] (Step S31) In step S31, the acquisition unit 31 acquires the taxonomy T2, the sentence set D2, and the auxiliary information F2. For example, the acquisition unit 31 may acquire the taxonomy T2, the sentence set D2, and the auxiliary information F2, part or all of them, from the memory of the information processing device 3, from another device connected via a network, or via an input device.

[0075] (Step S32) In step S32, the generation unit 32 refers to the sentence set D2 and generates a new phrase similar to the auxiliary information F2. Details of this step will be described later.

[0076] (Step S33) In step S33, the construction unit 33 estimates a relationship between the new phrase generated in step S32 and the existing phrases included in the taxonomy T2 that is consistent with the auxiliary information F2. The construction unit 33 also extends the taxonomy T2 by associating the new phrase with the existing phrase based on the estimation result. Details of this step will be described later.

[0077] <Details of step S32> Details of step S32 will be described with reference to Fig. 10. Fig. 10 is a flow diagram showing an example of a detailed flow of the process of step S32. As shown in Fig. 10, the process of step S32 includes steps S321 to S322.

[0078] (Step S321) In step S321, the generation unit 32 generates new candidate phrases by referring to the sentence set D2. For example, the generation unit 32 can use a language model in this step.

[0079] (Language Model) A language model is a model that outputs the probability of an input word string. For example, the probability of outputting the word string "his age is 100" when inputted into a certain language model is greater than the probability of outputting the word string "his birthday is 100" when inputted into the language model. This is because the language model holds the probability that the relationship between "he" and "100" is "age" and the probability that it is "birthday", and the probability that it is "age" is higher than the probability that it is "birthday". In addition, using the language model, it is possible to predict words that fit the masked parts of a word string in which some words are masked. As an example of a language model, for example, a language model generated by Reference 1 or Reference 2 below can be applied.

[0080] [References 1] Jacob Devlin et al., "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding", NAACL 2019. [Reference 2] Yinhan Liu et al., "RoBERTa: A Robustly Optimized BERT Pretraining Approach", arXiv 2019. For example, the generation unit 32 may extract new candidate phrases from the sentence set D2 using an extraction model generated using a language model. Here, the extraction model includes a language model and a classifier to which an output from the language model is input. The extraction model is generated by machine learning using teacher data that sets existing phrases appearing in the sentence set D2 as correct answers so that when a sentence included in the sentence set D2 is input to the language model, the classifier identifies and outputs a new phrase. In this case, the generation unit 32 sets phrases that are different from existing phrases among phrases extracted by inputting the sentence set D2 to the extraction model as new candidate phrases.

[0081] (Specific example of step S321) As a specific example, an example will be described in which the sentence set D2 includes the sentence "We bring you today's soccer game reports.... The two major baseball teams in New York are...." In this case, the generation unit 32 inputs the sentence set D2 into the extraction model to extract the phrases "soccer," "New York," and "baseball." Furthermore, the generation unit 32 sets the phrases "soccer" and "New York," which are different from the existing phrase "baseball," among the phrases extracted by the extraction model, as new candidate phrases.

[0082] (Step S322) In step S322, the generation unit 32 extracts phrases whose similarity to the auxiliary information F2 satisfies a predetermined condition from among the new candidate phrases, and sets them as new phrases to be added to the taxonomy T2. In other words, the generation unit 32 narrows down the new candidate phrases based on the condition that they are similar to the auxiliary information F2. One example of the predetermined condition is that the similarity between the phrase and the auxiliary information F2 is equal to or greater than a threshold. The similarity is as described above. However, the predetermined condition is not limited to this.

[0083] (Specific example of step S322) For example, an example will be described in which the auxiliary information F2 includes a natural language sentence "I would like to use this to recommend news distribution in Japan." In this case, the generation unit 32 calculates a similarity of 0.9 between the embedded representation of "soccer" and the embedded representation of the auxiliary information F2 among the new phrases obtained in step S321. Similarly, the generation unit 32 calculates a similarity of 0.1 between the embedded representation of "New York" and the embedded representation of the auxiliary information F2. In this way, a phrase that conforms to the auxiliary information F2 has a higher similarity to the auxiliary information F2 than a phrase that does not conform to the auxiliary information F2.

[0084] Here, for example, it is assumed that the threshold value is 0.7 In this case, the generation unit 32 extracts the phrase "soccer" whose similarity with the auxiliary information F2 is equal to or greater than the threshold value as a new phrase.

[0085] In this way, by narrowing down the search using auxiliary information F2, the phrase "soccer", which is similar to auxiliary information F2, is adopted as a new phrase to add to taxonomy T2, while "New York", which is not similar to auxiliary information F2, is not adopted.

[0086] 10, an example has been described in which new candidate phrases are generated in step S321, and then the candidates are narrowed down using the auxiliary information F2 in step S322. However, the present invention is not limited to this example, and the generation unit 32 may generate a new phrase similar to the auxiliary information F2 by referring to the auxiliary information F2 during the process of generating a new phrase from the sentence set D2.

[0087] <Details of step S33> Details of step S33 will be described with reference to Fig. 11. Fig. 11 is a flow diagram showing an example of a detailed flow of the process of step S33. As shown in Fig. 11, the process of step S33 includes steps S331 to S333.

[0088] (Step S331) In step S331, the constructing unit 33 calculates a score indicating the relationship between the new phrase and the existing phrase. For example, the constructing unit 33 can use a language model to calculate the score indicating the relationship between the new phrase and the existing phrase.

[0089] For example, the constructing unit 33 may calculate the likelihood of a sentence including a new phrase and an existing phrase using a language model, and may treat the calculated likelihood as a score indicating the above-mentioned relationship.

[0090] (Specific example of step S331) For example, a specific example of step S331 in the case where a new phrase "soccer" is generated in step S32 will be described. As an example, the construction unit 33 calculates 0.9 as a score between the new phrase "soccer" and the existing phrase "sports". This score is considered to indicate a relationship in which "soccer" is included in "sports". Furthermore, the construction unit 33 calculates 0.1 as a score between the new phrase "soccer" and the existing phrase "weather". This score indicates the possibility that these phrases have no relationship. Furthermore, the construction unit 33 calculates 0.8 as a score between the new phrase "soccer" and the existing phrase "baseball". This score is considered to indicate "a relationship in which both indicate types of sports".

[0091] (Step S332) In step S332, the construction unit 33 modifies the score indicating the relationship between the existing phrase and the new phrase by using the auxiliary information F2. For example, the construction unit 33 modifies the score by taking term1 as the existing phrase, term2 as the new phrase, and description as the auxiliary information F2 in the above-mentioned formula (1). However, the modification process using the auxiliary information F2 is not limited to this.

[0092] (Specific example of step S332) For example, the construction unit 33 obtains a similarity of 0.9 between the difference between the distributed representation of the new phrase "soccer" and the distributed representation of the existing phrase "sports" and the distributed representation of the auxiliary information F2. This indicates that the phrase "soccer" is included in "sports" and has a high similarity with the auxiliary information F2 "Want to use for recommending news distribution in Japan". Then, the construction unit 33 adds the similarity of 0.9 to the score of 0.9 between the phrases "soccer" and "sports" to revise the score to 1.8 according to formula (1).

[0093] Similarly, the construction unit 33 obtains a similarity of 0.1 between the difference between the embedded representation of the new phrase "soccer" and the embedded representation of the existing phrase "weather" and the embedded representation of the auxiliary information F2. This indicates that the relationship between the phrases "soccer" and "weather" and the similarity with the auxiliary information F2 are low. Then, the construction unit 33 adds the similarity of 0.1 to the score of 0.1 between the phrases "soccer" and "weather" to correct the score to 0.2 according to formula (1).

[0094] Similarly, the construction unit 33 obtains a similarity of 0.1 between the difference between the embedded representation of the new phrase "soccer" and the embedded representation of the existing phrase "baseball" and the embedded representation of the auxiliary information F2. This indicates that, for example, the relationship between the phrases "soccer" and "baseball" both indicating types of sports is low, and the similarity between the auxiliary information F2 "Want to use for recommending news distribution in Japan". Then, the construction unit 33 adds the similarity of 0.1 to the score of 0.8 between the phrases "soccer" and "baseball" to revise the score to 0.9 according to formula (1).

[0095] In this way, if a new phrase and an existing phrase have a "relationship in which both indicate a type of sport" that is inconsistent with auxiliary information F2, the score indicating such a relationship is revised to a relatively low value using auxiliary information F2.

[0096] (Step S333) In step S333, the construction unit 33 expands the taxonomy T2 based on the revised score. As an example, the construction unit 33 associates the new phrase with an existing phrase whose revised score between the new phrase and the existing phrase is equal to or greater than a threshold.

[0097] (Specific example of step S333) For example, the post-correction score between the existing phrase "sports" and the new phrase "soccer" is 1.8, which is greater than the threshold value of 1.0. Therefore, as shown in Fig. 11, the construction unit 33 associates the node N7 corresponding to the new phrase "soccer" with the node N3 corresponding to the existing phrase "sports" via an edge E6. This constructs a taxonomy T3 that is an extension of the taxonomy T2.

[0098] In this way, the new phrase "soccer" is associated with the existing phrase "sports" that has a "relationship between superordinate and subordinate concepts of news" that is consistent with the auxiliary information F2. The new phrase "soccer" is not associated with the existing phrase "baseball" that has a relationship with "both indicate types of sports" that is inconsistent with the auxiliary information F2.

[0099] <Advantages of this exemplary embodiment> As described above, the information processing device 3 according to this exemplary embodiment generates a phrase similar to the auxiliary information F2 as a new phrase to be added to the taxonomy T2 by referring to the sentence set D2. As a result, the new phrase to be added to the taxonomy T2 is more likely to be consistent with the auxiliary information F2. In addition, the information processing device 3 according to this exemplary embodiment estimates a relationship that is consistent with the auxiliary information F2 as a relationship between a new phrase and an existing phrase, and expands the taxonomy T2 based on the estimation result. As a result, the relationship included in the expanded taxonomy T3 is more likely to be consistent with the auxiliary information F2. Since the auxiliary information F2 is information related to the taxonomy T3 to be constructed, the present exemplary embodiment can construct a desired taxonomy T3 with greater accuracy.

[0100] <Another aspect 1 of this exemplary embodiment> In this exemplary embodiment, the auxiliary information F2 may not be used in either step S32 or S33. For example, the generation unit 32 may generate a new phrase from the sentence set D2 without using the auxiliary information F2 in step S32. Specifically, the generation unit 32 may omit the process of step S322 among steps S321 to S322 shown in Fig. 10. In this case, the generation unit 32 sets all new candidate phrases obtained in step S321 as new phrases to be added to the taxonomy T2.

[0101] Also, for example, in step S33, the construction unit 33 may extend the taxonomy T2 without using the auxiliary information F2. Specifically, the construction unit 33 may omit the process of step S332 among steps S331 to S333 shown in Fig. 11. However, the process of step S332 can be omitted only when the auxiliary information F2 is used in step S32. When the auxiliary information F2 is not used in step S32, the construction unit 33 executes the process of step S332 without omitting it.

[0102] In this way, even in aspect 1 where auxiliary information F2 is used in one of steps S32 and S33 but not in the other, it is possible to expand taxonomy T2 more accurately and construct the desired taxonomy T3 compared to a case where auxiliary information F2 is not used in both.

[0103] <Another aspect 2 of this exemplary embodiment> In the present exemplary embodiment, an example has been described in which the new candidate phrases generated by the generation unit 32 in step S321 are phrases extracted from the sentence set D2. However, the generation unit 32 may generate a new candidate phrase by predicting a phrase that is not necessarily included in the sentence set D2 with reference to the sentence set D2. In the process of predicting a new candidate phrase, for example, a language model can be used.

[0104] For example, the generation unit 32 masks a portion where an existing phrase appears in a sentence included in the sentence set D2, and predicts a phrase that matches the masked portion using a language model. In this case, the generation unit 32 sets the predicted phrase as a candidate new phrase.

[0105] <Another aspect 3 of this exemplary embodiment> In this exemplary embodiment, an example has been described in which a score indicating a relationship between a new phrase and an existing phrase is calculated using a language model in step S331. Note that the method of calculating a score indicating a relationship using a language model is not limited to this.

[0106] For example, the construction unit 33 may calculate a score indicating the relationship between an existing phrase and a new phrase using a discrimination model generated using a language model. In this case, the discrimination model includes a language model and a classifier to which an output from the language model is input. The discrimination model is trained so that when a sentence including an existing phrase and a new phrase is input to the language model, the classifier outputs a score indicating the relationship. The discrimination model is trained using teacher data in which the relationship between two existing phrases indicated by taxonomy T2 is regarded as a correct answer.

[0107] [Variation 1] In the above-described exemplary embodiments 2 and 3, the generation units 22 and 32 have been described as narrowing down the candidate phrases based on the similarity between the entire auxiliary information F1 and F2. The generation units 22 and 32 may narrow down the candidate phrases based on the similarity between the candidate phrases and a part of the auxiliary information F1 and F2, or a combination of the similarity between the different parts. For example, the generation unit 22 may narrow down the candidate phrases based on the similarity between the sentence (or phrase) having the highest similarity between the candidate phrase and the multiple sentences (or phrases) included in the auxiliary information F1 and F2. Also, for example, the generation unit 22 may narrow down the candidate phrases based on the average value of the similarity between each of the multiple sentences (or phrases) included in the auxiliary information F1 and F2 and the candidate phrase.

[0108] [Modification 2] In the above-described exemplary embodiments 2 and 3, the generation units 22 and 32 narrow down the candidate phrases based on the similarity to the distributed representations of the auxiliary information F1 and F2. However, the method of narrowing down the candidate phrases using the auxiliary information F1 and F2 is not limited to this. For example, a language model can be used in the process of narrowing down the candidate phrases using the auxiliary information F1 and F2.

[0109] Specifically, the generation units 22 and 32 may input sentences including candidate phrases and auxiliary information F1 and F2 to a language model, and narrow down the candidate phrases based on the certainty output from the language model.

[0110] [Modification 3] In the above-described exemplary embodiments 2 and 3, the constructing units 23 and 33 modify the scores indicating the relationships between a plurality of phrases based on the difference between the distributed representations indicating the relationships and the similarity between the auxiliary information F1 and F2. However, the method of estimating the relationships consistent with the auxiliary information F1 and F2 is not limited to this. For example, a language model can be used in the process of estimating the relationships consistent with the auxiliary information F1 and F2.

[0111] Specifically, the construction units 23, 33 may input a sentence including two phrases and auxiliary information F1, F2 into a language model, and calculate a score indicating a relationship consistent with the auxiliary information F1, F2 based on the accuracy output from the language model.

[0112] [Modification 4] In the above-described exemplary embodiments 2 and 3, the auxiliary information F1 and F2 include natural language sentences. However, the auxiliary information F1 and F2 do not have to include natural language sentences. For example, the auxiliary information F1 may be a set of words such as "cooking" and "ingredients," and the auxiliary information F2 may be a set of words such as "news," "Japan," "distribution," and "recommendation."

[0113] [Software implementation example] Some or all of the functions of the information processing devices 1, 2, and 3 may be realized by hardware such as an integrated circuit (IC chip), or may be realized by software.

[0114] In the latter case, the information processing devices 1, 2, and 3 are realized, for example, by a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter, referred to as computer C) is shown in FIG. 12. The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program P for operating the computer C as the information processing devices 1, 2, and 3. In the computer C, the processor C1 reads and executes the program P from the memory C2, thereby realizing each function of the information processing devices 1, 2, and 3.

[0115] The processor C1 may be, for example, a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a microcontroller, or a combination of these. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination of these.

[0116] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may further include a communication interface for transmitting and receiving data to and from other devices. The computer C may further include an input / output interface for connecting input / output devices such as a keyboard, a mouse, a display, and a printer.

[0117] Furthermore, the program P can be recorded on a non-transitory tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can obtain the program P via such a recording medium M. Furthermore, the program P can be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also obtain the program P via such a transmission medium.

[0118] [Additional Note 1] The present invention is not limited to the above-described embodiment, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining the technical means disclosed in the above-described embodiment are also included in the technical scope of the present invention.

[0119] [Additional Note 2] Some or all of the above-described embodiments can be described as follows. However, the present invention is not limited to the following described aspects.

[0120] (Appendix 1) An acquisition means for acquiring auxiliary information regarding the set of sentences and the taxonomy to be constructed; A generation means for generating phrases by referring to the set of sentences; A construction means for constructing the taxonomy by estimating relationships between a plurality of phrases including the phrase, using the auxiliary information in one or both of the generating means and the constructing means; Information processing device.

[0121] According to the above configuration, when auxiliary information is used in the generating means, the phrases included in the constructed taxonomy are more likely to be the desired phrases that are in line with the content of the auxiliary information. Also, when auxiliary information is used in the constructing means, the relationships between multiple phrases in the constructed taxonomy are more likely to be the desired relationships that are in line with the content of the auxiliary information. Therefore, the desired taxonomy can be constructed with greater accuracy.

[0122] (Appendix 2) The information processing device according to claim 1, wherein the generation means generates the phrase similar to the auxiliary information by referring to the set of sentences.

[0123] According to the above configuration, since a phrase contained in the constructed taxonomy is similar to the auxiliary information, the phrase is more likely to be the desired one.

[0124] (Appendix 3) The information processing device described in Appendix 2, wherein the generation means generates candidate phrases by referring to the collection of sentences, and extracts, from the candidate phrases, phrases whose similarity to the auxiliary information satisfies a predetermined condition.

[0125] According to the above configuration, the possibility that undesired phrases will be included in the constructed taxonomy can be reduced.

[0126] (Appendix 4) 4. The information processing device according to claim 1, wherein the construction means estimates a relationship between the plurality of phrases that is consistent with the auxiliary information.

[0127] According to the above configuration, the relationships contained in the constructed taxonomy are consistent with the auxiliary information, so that the relationships are more likely to be the desired ones.

[0128] (Appendix 5) The information processing device described in Appendix 4, wherein the construction means calculates information indicating the relationship between the multiple phrases, and estimates a relationship that is consistent with the auxiliary information by modifying the calculated information using the auxiliary information.

[0129] According to the above configuration, the relationships included in the constructed taxonomy are relationships that have been corrected using auxiliary information, so there is a high possibility that the relationships are the desired ones.

[0130] (Appendix 6) 6. The information processing device according to claim 1, wherein the auxiliary information includes a natural language sentence.

[0131] According to the above configuration, it is possible to use a natural language sentence that can be easily generated as auxiliary information used to generate a more accurate taxonomy.

[0132] (Appendix 7) 6. The information processing device according to claim 1, wherein the auxiliary information includes a set of words.

[0133] According to the above configuration, a set of words that can be easily generated can be used as auxiliary information used to generate a more accurate taxonomy.

[0134] (Appendix 8) The acquiring means further acquires a partial taxonomy constituting a part of the taxonomy to be constructed, The generating means generates new phrases different from phrases included in the partial taxonomy by referring to the set of sentences; The construction means extends the partial taxonomy by estimating a relationship between a phrase included in the partial taxonomy and the new phrase. 8. An information processing device according to any one of appendix 1 to 7.

[0135] According to the above configuration, a taxonomy constructed by extending a partial taxonomy can be made to be a desired one that more accurately matches the content of the auxiliary information.

[0136] (Appendix 9) Obtaining auxiliary information about the set of sentences and the taxonomy to be constructed; generating phrases by referencing the set of sentences; and constructing the taxonomy by inferring relationships between a plurality of phrases that include the phrase; An information processing method that uses the auxiliary information in one or both of generating the phrases and building the taxonomy.

[0137] According to the above configuration, the same effects as those of Supplementary Note 1 are achieved.

[0138] (Appendix 10) A program for causing a computer to function as an information processing device, the program comprising: An acquisition means for acquiring auxiliary information regarding the set of sentences and the taxonomy to be constructed; A generation means for generating phrases by referring to the set of sentences; a construction means for constructing the taxonomy by estimating relationships between a plurality of phrases including the phrase; A program that uses the auxiliary information in one or both of the generating means and the constructing means.

[0139] According to the above configuration, the same effects as those of Supplementary Note 1 are achieved.

[0140] [Additional Note 3] A part or all of the above-described embodiments can be further expressed as follows.

[0141] At least one processor, the processor comprising: an acquisition process for acquiring auxiliary information about the set of sentences and the taxonomy to be constructed; A generation process of generating phrases by referring to the set of sentences; A construction process for constructing the taxonomy by estimating relationships between a plurality of phrases including the phrase; An information processing device that uses the auxiliary information in one or both of the generating process and the constructing process.

[0142] The information processing device may further include a memory that stores a program for causing the processor to execute the acquisition process, the generation process, and the construction process. The program may be recorded on a computer-readable, non-transitory, tangible recording medium. [Explanation of symbols]

[0143] 1, 2, 3 Information processing device 11, 21, 31 Acquisition part 12, 22, 32 generation section 13, 23, 33 Construction Department

Claims

1. An acquisition means for acquiring auxiliary information regarding the set of sentences and the taxonomy to be constructed; A generation means for generating phrases by referring to the set of sentences; A construction means for constructing the taxonomy by estimating relationships between a plurality of phrases including the phrase, using the auxiliary information in one or both of the generating means and the constructing means; The auxiliary information includes a natural language sentence. Information processing device.

2. The information processing apparatus according to claim 1 , wherein the generating means generates the phrase similar to the auxiliary information by referring to the set of sentences.

3. The information processing apparatus according to claim 2 , wherein the generating means generates candidate phrases by referring to the set of sentences, and extracts, from the candidate phrases, phrases whose similarity to the auxiliary information satisfies a predetermined condition.

4. The information processing apparatus according to claim 1 , wherein the constructing means estimates a relationship between the plurality of phrases that is consistent with the auxiliary information.

5. The information processing device according to claim 4 , wherein the construction means calculates information indicating a relationship between the plurality of phrases, and estimates a relationship that is consistent with the auxiliary information by correcting the calculated information using the auxiliary information.

6. The information processing apparatus according to claim 1 , wherein the auxiliary information includes a set of words.

7. The acquiring means further acquires a partial taxonomy constituting a part of the taxonomy to be constructed, The generating means generates new phrases different from phrases included in the partial taxonomy by referring to the set of sentences; The construction means extends the partial taxonomy by estimating a relationship between a phrase included in the partial taxonomy and the new phrase. The information processing device according to claim 1 .

8. At least one processor obtains auxiliary information about the set of sentences and the taxonomy to be constructed; generating phrases by referring to the set of sentences, The at least one processor constructs the taxonomy by inferring relationships between a plurality of phrases that include the phrase; the at least one processor uses the auxiliary information in generating the phrases and / or building the taxonomy; The auxiliary information includes a natural language sentence. Information processing methods.

9. A program for causing a computer to function as an information processing device, the program comprising: An acquisition means for acquiring auxiliary information regarding the set of sentences and the taxonomy to be constructed; A generation means for generating phrases by referring to the set of sentences; a construction means for constructing the taxonomy by estimating relationships between a plurality of phrases including the phrase; using the auxiliary information in one or both of the generating means and the constructing means; The auxiliary information includes a natural language sentence. program.

10. An acquisition means for acquiring auxiliary information regarding a set of sentences and a taxonomy to be constructed; A generation means for generating phrases by referring to the set of sentences; A construction means for constructing the taxonomy by estimating relationships between a plurality of phrases including the phrase, using the auxiliary information in one or both of the generating means and the constructing means; the generating means generates the phrase similar to the auxiliary information by referring to the set of sentences. Information processing device.

Images