Analysis device, analysis method and storage medium
By using the calculation unit, the application unit and the decision unit in the analysis device to calculate the vocabulary similarity and determine the relationship between vocabulary and category, the problem of inaccurate vocabulary classification in the prior art is solved, and more efficient vocabulary and category association and accuracy are achieved. Search results.
Patent Information
- Application Number
- CN202010122489.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2019-09-06
- Filing Date
- 2020-02-27
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2040-02-27
AI Technical Summary
The prior art cannot effectively determine the association between vocabulary and categories when using a similarity calculator for vocabulary classification, resulting in undesirable results and inaccurate keyword allocation, increasing development costs and complexity.
Analytical device is designed, including a calculation unit, an application unit, a decision unit and a display control unit. By calculating the similarity between vocabulary, vocabulary with high similarity is applied to template statements, determining the category of vocabulary, and displaying the association in the display unit. so that users can more easily grasp the relationship between vocabulary and categories.
Achieve more accurate and efficient vocabulary and category associations, reduce undesired keyword allocation, reduce development costs, and improve the accuracy of search results.
Smart Images

Figure CN112463934B_ABST
Abstract
Description
Technical Field
[0001] Embodiments of the present invention relate to an analysis device, an analysis method, and a storage medium. Background Art
[0002] In searches performed by inputting natural sentences, it is necessary to collect the words (search keywords) that the user will input in advance. In the past, developers manually assigned corresponding search keywords to each content. Therefore, it is necessary to consider what search keywords to use for each content, and it costs money to generate a database. Although a keyword can be expanded using a synonym dictionary, other synonyms need to be manually assigned. In recent years, similarity calculator technology has been used, which uses a large amount of documents to learn the similarity model between words and calculate the similarity of the meanings between words. Summary of the invention
[0003] However, in the prior art, it is not possible to confirm in advance what keywords are assigned to each category. A similarity calculator generated by machine learning using a large number of documents sometimes gives undesirable results as similarities in meaning, and sometimes assigns words (keywords) that developers do not want to a category. The problem to be solved by the present invention is to provide an analysis device, analysis method, and program that can more easily grasp the words associated with a category.
[0004] The analysis device of the embodiment includes a calculation unit, an application unit, a determination unit, and a display control unit. The calculation unit calculates a first similarity between a first word indicating a category and a second word. The application unit applies one or more second words whose first similarity is greater than a first threshold value to one or more template sentences. The determination unit analyzes the template sentence containing the second word and classifies the second word into one or more first categories. The display control unit displays the second word used in the classification of the first category on the display unit for each of the first categories.
[0005] According to the above-described estimation device, it is possible to more easily grasp words associated with categories. BRIEF DESCRIPTION OF THE DRAWINGS
[0006] Figure 1 This is a block diagram showing an example of the functional configuration of the analysis device according to the first embodiment.
[0007] Figure 2 This is a block diagram showing an example of the functional configuration of the dialogue system according to the first embodiment.
[0008] Figure 3 This is a flowchart showing an example of the voice dialogue process according to the first embodiment.
[0009] Figure 4AThis is a diagram showing a calculation example 1 of the degree of relevance between words and categories according to the first embodiment.
[0010] Figure 4B This is a diagram showing a second calculation example of the degree of relevance between words and categories according to the first embodiment.
[0011] Figure 5 This is a diagram showing Example 1 of display information indicating association between words and categories according to the first embodiment.
[0012] Figure 6 This is a diagram showing a second example of display information indicating the association between words and categories according to the first embodiment.
[0013] Figure 7 This is a flowchart showing an example of a method for calculating the degree of relevance between words and categories according to the first embodiment.
[0014] Figure 8 This is a diagram showing an example of display information of a change in acceptance type according to the second embodiment.
[0015] Fig. 9 This is a diagram showing an example of display information of deletion of the acceptance category according to the second embodiment.
[0016] Fig.10 This is a diagram showing an example of additional display information of the acceptance category according to the second embodiment.
[0017] Fig.11 This is a diagram showing an example of display information for accepting a change in the degree of relevance between a word and a category according to the third embodiment.
[0018] Fig.12 This is a diagram showing an example of display information indicating the association between words and categories according to the fifth embodiment.
[0019] Fig.13 It is a diagram showing an example of the hardware configuration of the analysis device according to the first to fifth embodiments.
[0020] Description of symbols
[0021] 1: Calculation department;
[0022] 2: Applicable part;
[0023] 3: Decision Department;
[0024] 4: Display control unit;
[0025] 5: Search department;
[0026] 10: Analytical device;
[0027] 20: Dialogue system;
[0028] 101: storage unit;
[0029] 301: control device;
[0030] 302: main storage device;
[0031] 303: auxiliary storage device;
[0032] 304: display device;
[0033] 305: input device;
[0034] 306: Communication device;
[0035] 310: Bus. DETAILED DESCRIPTION
[0036] Hereinafter, embodiments of the analysis device, the analysis method, and the storage medium will be described in detail with reference to the accompanying drawings.
[0037] [Example of functional configuration of analysis device]
[0038] Figure 1 1 is a block diagram showing an example of the functional configuration of the analysis device 10 according to the first embodiment. The analysis device 10 according to the first embodiment includes a calculation unit 1 , an application unit 2 , a determination unit 3 , a display control unit 4 , and a storage unit 101 .
[0039] The storage unit 101 stores information. The storage unit 101 stores, for example, content classified by category (content database). Categories include, for example, types of merchandise such as clothing, bags, and accessories, types of cuisine such as Japanese food, Chinese food, and Western food, target users such as female and male, and price ranges such as cheap and expensive. Contents include, for example, stores that sell goods. One or more categories are registered in the content.
[0040] The calculation unit 1 calculates the similarity (first similarity) between the vocabulary (first vocabulary) representing the category and the vocabulary (second vocabulary). In addition, the method of accepting the vocabulary representing the category can be arbitrary. For example, the first vocabulary can be directly input by the user, or the user can select from a plurality of first vocabularies.
[0041] In addition, for example, a vocabulary (first vocabulary) indicating a category may be obtained from input information such as speech data. When receiving speech data, the calculation unit 1 calculates the similarity (second similarity) between the vocabulary (third vocabulary) included in the speech data and the vocabulary (first vocabulary) indicating a category. Furthermore, the calculation unit 1 calculates the similarity (first similarity) between a vocabulary (first vocabulary) whose similarity (second similarity) is greater than a threshold value (second threshold value) and a vocabulary (second vocabulary). Specifically, when the input information is, for example, a user's speech "I want to buy a guitar", the vocabulary "musical instrument" of the category is obtained as a vocabulary whose similarity (second similarity) to "guitar" is greater than the threshold value (second threshold value). In this case, the calculation unit 1 calculates the similarity (first similarity) between the vocabulary "musical instrument" and the vocabulary (second vocabulary).
[0042] Specifically, the calculation unit 1 calculates the similarity of the meanings between words. The specific method of calculating the similarity can be arbitrary. For example, the more similar the usage methods in the document are, the higher the similarity between the words calculated by the calculation unit 1. In addition, the calculation unit 1 can also output words with high similarity to a certain word (words with similarity greater than a threshold value).
[0043] The application unit 2 inputs a vocabulary (first vocabulary) representing a category to the calculation unit 1, and receives one or more vocabulary (second vocabulary) whose similarity (first similarity) is greater than a threshold value (first threshold value) from the calculation unit 1. For example, when the category is "musical instrument", the application unit 2 receives "guitar", "piano", "clarinet" and "harp" from the calculation unit 1. Furthermore, the application unit 2 applies the one or more vocabulary received from the calculation unit 1 to one or more template sentences. The template sentence, for example, represents a speech sentence for searching for content. The template sentence is generated, for example, based on a large number of speech sentences collected in advance. Specifically, the template sentences are "I want to buy [musical instrument]", "I want [musical instrument]", "[musical instrument] store" and "[musical instrument]". The part of [musical instrument] can apply the vocabulary "guitar", "piano", "clarinet" and "harp", for example.
[0044] The decision unit 3 analyzes the template sentence including the vocabulary (second vocabulary) and classifies the vocabulary into one or more categories (first categories). Specifically, the decision unit 3 analyzes "I want to buy a guitar" and classifies "guitar" into the category "musical instrument".
[0045] The specific parsing method of the template sentence (speech sentence) can be arbitrary. For example, a large number of speech sentences used to retrieve content are collected in advance, and a label is attached to which category each speech sentence has mentioned, and it is used as learning data to create a model. Several learning methods can be considered, but the following example is explained. First, when each speech sentence included in the learning data is a language in which the vocabulary is not segmented (such as Japanese, etc.), morpheme analysis is performed to segment the speech sentence into vocabulary. Next, the vocabulary of the category registered for the content is obtained. Next, the calculation unit 1 is used to calculate the similarity between the vocabulary in the speech sentence and the vocabulary of the category. Next, the model is learned using the vocabulary in the speech sentence and the calculated similarity as feature quantities and the category marked with the label as the correct value. In addition, when the actual analysis is performed, the decision unit 3 also classifies the vocabulary into categories by calculating the similarity between the vocabulary in the template sentence to which the vocabulary is applied and the category, as described above.
[0046] In addition, since the calculation unit 1 only calculates the similarity between words, it is impossible to determine which keyword corresponds to which category using only the calculation unit 1. In fact, the association between keywords and categories is not determined only by the similarity, and sometimes the association changes depending on the wording of the utterance sentence depending on the design of the dialogue system. For example, the utterance sentence "I want to buy children's clothing" corresponds to the category "for children", and the utterance sentence "places where children can be brought in" corresponds to the category "for families".
[0047] Therefore, the determination unit 3 is created based on the similarity calculated by the calculation unit 1. The determination unit 3 collects a large number of utterances for retrieval in advance, and learns a model that determines which category a keyword in the utterance corresponds to using the utterances and the calculation unit 1. Since the determination unit 3 is using the learning model, it is not possible to directly confirm which keyword corresponds to which category based on the model. In addition, since the machine learning is performed based on a large number of utterances, it is possible that undesirable keywords are associated with categories. Therefore, for example, the developer needs to confirm and change undesirable keywords in advance.
[0048] The display control unit 4 displays the words (second words) used in the classification of the category (first category) on the display unit for each category (first category). The display unit is, for example, a display device such as a liquid crystal display. The display unit may be provided in the analysis device 10 or in an external device.
[0049] The database storing the contents is prepared for each domain of the interactive system, for example. For example, in the case of store search, the domain is divided for each department and shopping hall including a plurality of stores. In addition, for example, the domain may be divided for each search object such as store search, product search, and service search.
[0050] The calculation unit 1 and the determination unit 3 described above may be customized for each domain, or may be used commonly for all domains.
[0051] Here, a description will be given of a voice dialogue process when a user searches for content using the dialogue system using the functions (calculation unit 1, determination unit 3, and storage unit 101) of the analysis device 10.
[0052] [Example of the functional configuration of a dialogue system]
[0053] Figure 2 1 is a block diagram showing an example of the functional configuration of the dialog system 20 according to the first embodiment. The dialog system 20 according to the first embodiment includes a calculation unit 1, a determination unit 3, a search unit 5, and a storage unit 101. Figure 1 Same, so omitted.
[0054] The search unit 5 searches for content based on the category (first category) classified by the determination unit 3. Figure 3 The details of the content search method are described.
[0055] In addition, the analysis device 10 and the dialogue system 20 may be implemented by the same device. In the embodiment, for the sake of explanation, when explaining the operation of the analysis process, the analysis device 10 is referred to, and when explaining the operation of the speech dialogue process, the dialogue system 20 is referred to.
[0056] Figure 3 This is a flowchart showing an example of the voice dialogue process according to the first embodiment. Figure 3 The example shows a case where the user's spoken input is "I want to buy a guitar".
[0057] First, the utterance sentence is analyzed (step S1 ). Specifically, through the processing of the calculation unit 1 and the determination unit 3 described above, the category “musical instrument” is output.
[0058] Next, the search unit 5 creates one or more search queries based on the category "musical instrument" obtained by the process of step S1 (step S2). The search query includes, for example, words (search keywords) whose relevance to the category "musical instrument" is greater than a threshold. The relevance will be described later.
[0059] Next, the search unit 5 searches for content using the search query created by the process of step S2 (step S3). Next, the search unit 5 determines whether the search result is 0 (step S4). If the search result is 0 (step S4: yes), the search result is notified (step S5), and the process ends.
[0060] When the search result is not 0 (step S4: No), the search unit 5 determines whether the search result is less than X (threshold value) (step S6). When the search result is more than X (step S6: No), the search unit 5 requests other conditions (step S7). When the search result is less than X (step S6: Yes), the search unit 5 displays the search result on the display unit.
[0061] In making the above Figure 2 In the case of a new domain of the dialogue system 20, a database storing content (content DB) is first created. In addition, when the calculation unit 1, the database storing template sentences (template sentence DB), and the determination unit 3 are used in common for the domain, an existing database can be used.
[0062] When the calculation unit 1 , the template sentence DB, and the determination unit 3 are stored for each domain, a large number of utterances for searching the content DB are collected, and the calculation unit 1 , the template sentence DB, and the determination unit 3 are created based on the utterances.
[0063] When the functional blocks of the analysis device 10 and the dialogue system 20 are used in common for domains, when creating a new domain, only the content DB needs to be designed. In addition, since there is no need to consider synonyms and similar words of the category, only the category needs to be registered in the content. However, in this state, the user does not actually know what search keyword to use to search for each content. For example, a word with a higher degree of relevance to the category than a set threshold is used as the search keyword.
[0064] Therefore, the determination unit 3 determines the degree of association between the category (first category) and the vocabulary (second vocabulary) based on the number of template sentences used when classifying the vocabulary into categories. And the display control unit 4 displays the vocabulary with the degree of association. Thus, the user can more easily grasp the vocabulary associated with the category.
[0065] Figure 4A 1 is a diagram showing a calculation example 1 of the degree of association between words and categories in the first embodiment. The application unit 2 inputs the category "musical instrument" to the calculation unit 1, and applies one or more words (second words) received from the calculation unit 1 to one or more template sentences. Figure 4A In the example of , “guitar”, “piano”, “clarinet”, and “harp” are exemplified as one or more words received from the calculation unit 1 .
[0066] The decision unit 3 analyzes the template sentence including the vocabulary (second vocabulary) and classifies the vocabulary into a category (first category). Figure 4AIn the example of , the decision unit 3 analyzes "want to buy a guitar" and classifies "guitar" into the "musical instrument" category. In addition, the decision unit 3 analyzes "want a guitar" and classifies "guitar" into the "musical instrument" category. In addition, the decision unit 3 analyzes "guitar store" and classifies "guitar" into the "musical instrument" category. In addition, the decision unit 3 analyzes "guitar" and classifies "guitar" into the "musical instrument" category.
[0067] Next, the decision unit 3 decides the degree of relevance between the category (first category) and the vocabulary (second vocabulary) based on the number of template sentences used when classifying the vocabulary (second vocabulary) into categories. Figure 4A In the example of , in all template sentences to which "guitar" is applied, "guitar" is classified into the "musical instrument" category. Therefore, the decision unit 3 decides the degree of association between the "musical instrument" category and "guitar" to be 1.0.
[0068] Figure 4B 1 is a diagram showing a calculation example 2 of the degree of association between words and categories in the first embodiment. The application unit 2 inputs the category "musical instrument" to the calculation unit 1, and applies one or more words (second words) received from the calculation unit 1 to one or more template sentences. Figure 4B In the example of , “guitar”, “piano”, “clarinet”, and “harp” are exemplified as one or more words received from the calculation unit 1 .
[0069] The decision unit 3 analyzes the template sentence including the vocabulary (second vocabulary) and classifies the vocabulary into a category (first category). Figure 4B In the example of , the decision unit 3 analyzes “want to buy a recorder” and classifies “recorder” into the “musical instrument” category. In addition, the decision unit 3 analyzes “want a recorder” and classifies “recorder” into the “musical instrument” category. In addition, the decision unit 3 analyzes “recorder store” and classifies “recorder” into the “musical instrument” category. In addition, the decision unit 3 analyzes “recorder” and classifies “recorder” into the “children’s product” category.
[0070] Next, the decision unit 3 decides the degree of relevance between the category (first category) and the vocabulary (second vocabulary) based on the number of template sentences used when classifying the vocabulary (second vocabulary) into categories. Figure 4B In the example, in the three template sentences to which "clarinet" is applied, "clarinet" is classified into the "musical instrument" category. In addition, in one template sentence to which "clarinet" is applied, "clarinet" is classified into the "children's product" category. Therefore, the decision unit 3 determines the degree of association between the "musical instrument" category and "clarinet" to be 0.75, and determines the degree of association between the "children's product" category and "clarinet" to be 0.25.
[0071] Figure 5This is a diagram showing Example 1 of display information indicating association between words and categories according to the first embodiment. Figure 5 The example of shows how to visualize the association between words and categories in a list form. Figure 5 In the search results, words related to each category are displayed as search keywords. A search keyword related to a plurality of categories is given a degree of relevance.
[0072] Figure 6 This is a diagram showing a second example of display information indicating the association between words and categories according to the first embodiment. Figure 6 The example of shows how to visualize the association between words and categories in the form of a map. Figure 6 In , words related to each category are displayed around each category. The position of the word is determined by the relevance between the word and the category, and the similarity between the words. Words related to multiple categories are linked to the categories with multiple arrows, and the arrows are given a relevance. Figure 6 In the example of , the similarity between words (search keywords) can also be visualized by the distance between words.
[0073] Figure 7 This is a flowchart showing an example of a method for calculating the degree of association between a word and a category according to the first embodiment. First, the application unit 2 inputs a word (first word) representing a category into the calculation unit 1, and obtains one or more words (second words) from the calculation unit 1 (step S11). Next, the application unit 2 applies the one or more words (second words) obtained by the process of step S11 to one or more template sentences (step S12).
[0074] Next, the decision unit 3 analyzes the template sentence including the vocabulary (second vocabulary) and classifies the vocabulary into a category (first category) (step S13). Next, the decision unit 3 determines the degree of association between the category (first category) and the vocabulary based on the number of template sentences used when classifying the vocabulary (second vocabulary) into categories (step S14).
[0075] Next, the display control unit 4 displays the vocabulary (second vocabulary) used for classification of the category (first category) on the display unit for each category (first category) (step S15 ). Figure 7 The example shows a case where, when the degree of relevance is 1.0, the display of the degree of relevance is omitted, and when a word is related to two or more categories, the word is displayed with the degree of relevance added thereto.
[0076] As described above, in the analysis device 10 of the first embodiment, the calculation unit 1 calculates the similarity (first similarity) between the vocabulary (first vocabulary) indicating the category and the vocabulary (second vocabulary). The application unit 2 applies one or more second vocabulary having a first similarity greater than a first threshold value to one or more template sentences. The determination unit 3 analyzes the template sentence including the second vocabulary and classifies the second vocabulary into one or more categories (first categories). Furthermore, the display control unit 4 displays the second vocabulary used in the classification of the first category on the display unit for each first category.
[0077] As a result, according to the analysis device 10 of the first embodiment, it is possible to more easily grasp words associated with categories.
[0078] (Second embodiment)
[0079] Next, a second embodiment will be described. In the description of the second embodiment, the same description as that of the first embodiment will be omitted.
[0080] pass Figure 5 and Figure 6 The result after confirming the words associated with the category is sometimes different from what the user wants. For example, due to the inclusion of an incorrect value in the setting of the calculation unit 1, words that are not close in meaning are sometimes associated with the category. In addition, for example, in the case of searching for stores, even in stores classified as the category "musical instruments", it is not desirable to search for "piano" in the case of stores that do not sell "pianos". Therefore, in the second embodiment, a configuration is described for enabling the association between categories and words to be changed when the association results in an undesirable result.
[0081] Figures 8 to 11 This is an example of display information (user interface) that accepts changes, deletions, or additions to categories (first categories) associated with a word (second word). Figure 5 and Figure 6 The visualized search keyword may be used to describe a new search keyword. The category item can be selected from the list of categories in the current content DB. When receiving the input of the search keyword item, the display control unit 4 displays the currently corresponding category in the category item.
[0082] Figure 8 This is a diagram showing an example of display information for accepting a change in category according to the second embodiment. For example, when "sportswear" in the displayed category item is changed to "clothing" and a press of the "register" button is accepted, the display control unit 4 reflects the change content.
[0083] Fig. 9 This is a diagram showing an example of display information for accepting deletion of a category according to Embodiment 2. For example, when receiving a click of a trash can icon adjacent to the category "musical instrument" of the search keyword "piano", the display control unit 4 reflects deletion of the word "piano" associated with the category "musical instrument".
[0084] Fig.10 This is a diagram showing an example of display information for accepting the addition of a category according to the second embodiment. For example, when "cake" is added as a category associated with the search keyword "Sabre" and a "register" button is accepted, the display control unit 4 reflects the addition of the word "Sabre" to the category "cake".
[0085] Even if the user who performs the modification, deletion, and addition operation does not understand the internal mechanism of the system of the analysis device 10, he can observe Figure 5 and Figure 6 , for undesired search keywords, only operate Figures 8 to 11 Such a simple user interface makes it the desired design.
[0086] Next, a method of reflecting the change, deletion, and addition of a category will be described. For example, three methods of reflecting the change, deletion, and addition of a category are considered.
[0087] The first method is a method of changing the operation of the calculation unit 1. The calculation unit 1 changes the similarity according to the change, deletion or addition of the category (first category) associated with the vocabulary (second vocabulary). Specifically, for example, when the search keyword "sweater" is to be removed from the category "sportswear", the similarity between the vocabulary associated with the category "sportswear" and the search keyword "sweater" is reduced (set to be less than a threshold value). When the search keyword "sweater" is to be added to the category "clothing", the similarity between the vocabulary associated with the category "clothing" and the search keyword "sweater" is increased (set to be above a threshold value).
[0088] The second method is to change the operation of the decision unit 3. The decision unit 3 changes, deletes or adds the category to which the vocabulary (the second vocabulary) is classified according to the change, deletion or addition of the category (the first category) associated with the vocabulary. Specifically, the decision unit 3 uses a learning model, but instead of outputting the judgment result of the learning model as it is, a process of changing the judgment result of the learning model according to a rule is added. By changing the rule, the search keyword can be changed. For example, in the case where you want to remove the search keyword "sweater" from the category "sportswear", even if the learning model outputs the category "sportswear", a rule for removing the category "sportswear" is recorded. In addition, for example, in the case where you want to add the search keyword "sweater" to the category "clothing", even if the learning model does not output the category "clothing", a rule for adding the category "clothing" is recorded.
[0089] The third method is to change the content DB. The decision unit 3 changes, deletes or adds the category (second category) registered in the content by the content DB according to the change, deletion or addition of the category (first category) associated with the vocabulary (second vocabulary). Specifically, for example, the search keyword to be changed is newly set as a category. For example, in the case where the search keyword "sweater" is to be changed, the category "sweater" is created. The search keyword "sweater" corresponds to the category "sweater" with the highest similarity, so it no longer corresponds to other categories. In addition, for example, in the case where the search keyword "sweater" is to be removed from the category "sportswear", the category "sweater" is created. In addition, for example, in the case where the search keyword "sweater" is to be added to the category "clothing", the category "sweater" is created and the category "sweater" is registered in the content registered in the category "clothing". When the dialogue system 20 is executed, the search keyword "sweater" is not regarded as the category "clothing", but because the content registered in the category "clothing" and the content registered in the category "sweater" are the same, even if the category "sweater" is used for searching, the search results are the same.
[0090] When a search keyword corresponds to multiple categories, it is impossible to know which category it is determined to be based on the utterance. Therefore, sometimes you want to change it to correspond to only one category. For example, Figure 5 and Figure 6 In the example, the search keyword "T-shirt" is included in both the category "sportswear" and the category "clothing". When changing to correspond only to the category "clothing", the search keyword "T-shirt" can be removed from the category "sportswear" by any of the above three methods.
[0091] On the contrary, sometimes you want to match a search keyword to multiple categories. Figure 5 and Figure 6 In the example, the search keyword "baked desserts" corresponds to the two categories "cake" and "cookies". However, when a speech sentence including "baked desserts" is input, the two categories "cake" and "cookies" are not output, but the category "cake" or the category "cookies" is output according to the wording of the speech sentence. Sometimes, it is desired to output both the categories "cake" and "cookies" regardless of the wording of the speech sentence. Even if the calculation unit 1 is changed, the two categories cannot be output, so the determination unit 3 or the content DB is changed.
[0092] In the case of the change decision unit 3, the latter rule processing part is described in such a way that the search keyword "baked desserts" outputs both the categories "cake" and "cookies". In the case of the change content DB, the category "baked desserts" is created, and the category "baked desserts" is registered for all contents registered in the categories "cake" and "cookies". In this way, when the search keyword "baked desserts" is input, the category "baked desserts" is output, and the contents registered in the category "baked desserts" are matched.
[0093] exist Figure 5 and Figure 6 In the initial state, when multiple categories are supported, the output category is different according to the wording of the spoken sentence. If the user does not make any special changes, it is registered in the original state.
[0094] In the user Fig.10 When the search keyword is associated with a plurality of categories, the plurality of categories are output regardless of the utterance. The following specifically describes the processing when a plurality of categories are output when performing a search, taking two methods as examples.
[0095] The first is to narrow (limit) the method to one category.
[0096] User: I want to buy baked snacks
[0097] Dialogue system: Do you prefer cookies or cake?
[0098] User: I like cookies
[0099] Retrieval query: select *, from database where category LIKE "cookies"
[0100] Since the search keyword "baked confectionery" is associated with the categories "cookies" and "cake", the determination unit 3 outputs both categories. The interactive system 20 does not perform a search at this stage but asks the user which category he / she likes, and performs a search after narrowing the category to one.
[0101] The second is a method that directly retrieves multiple categories.
[0102] User: I want to buy baked snacks
[0103] Search query: select * from database where category LIKE "cookies" OR category LIKE "cake"
[0104] Dialogue System: Found 20 items for you. Do you have a preferred price range?
[0105] User: Cheaper is better
[0106] Search query: select * from database where (category LIKE "cookies" OR category LIKE "cake") AND category LIKE "low price"
[0107] System: Found 5 items for you
[0108] The dialog system 20 performs an OR search of the categories "cookies" and "cakes". After that, it searches for content using other limiting conditions.
[0109] In the above example, when the user says "baked snacks", the user may not have decided which one is better, "cookies" or "cakes". The first method mentioned above is a method of letting the user decide which one. The second method mentioned above is a method of directly searching in a vague state. Regarding the first method, since there are few search results, it has the advantage of quickly limiting to a small number of items. Regarding the second method, it has the following advantage, that is, when the user cannot decide whether it is "cookies" or "cakes", it can be limited by other conditions without forcing the user to decide which one.
[0110] (Third embodiment)
[0111] Next, a third embodiment will be described. In the description of the third embodiment, the same description as that of the second embodiment will be omitted. In the third embodiment, a case where a change of a vocabulary associated with a category is accepted by a more direct operation will be described.
[0112] Fig.11 This is a diagram showing an example of display information for accepting a change in the degree of relevance between a word and a category according to the third embodiment. Fig.11 The display format of the display information is a mapping format in which the association between the category (first category) and the vocabulary (second vocabulary) is represented by the position of the category and the position of the vocabulary. The display control unit 4 receives the change of the category associated with the vocabulary by input indicating the change of the vocabulary position. And the calculation unit 1 changes the similarity according to the change of the vocabulary position.
[0113] Specifically, the user can drag a word (search keyword) and place it at any position. The calculation unit 1 updates the similarity between displayed words based on the position of the word. Furthermore, the determination unit 3 recalculates the relevance between the word and the category, and the display control unit 4 updates the display information.
[0114] In the second embodiment described Figures 8 to 10 In the method, in order to change the association between words and categories, the similarity calculation process based on the calculation unit 1 is also changed, but sometimes the similarity does not match what the user wants. Fig.11In the method, since the position of the word (search keyword) can be freely changed, the similarity between words can be changed to a desired value according to the position of the word. If the value of the similarity is set to a preferred value, for example, when the calculation unit 1 is used in other domains in the future, there is no need to adjust the operation of the calculation unit 1 again.
[0115] (Fourth embodiment)
[0116] Next, the fourth embodiment will be described. In the description of the fourth embodiment, the same description as the second embodiment will be omitted. In the fourth embodiment, Example 1 is described in which a change is made to words associated with a category in common for domains and the change is made for each domain.
[0117] As reasons for changing the association between a word (search keyword) and a category, for example, two reasons can be considered. One is that an error occurs in the work of the calculation unit 1. The other is that the corresponding search keyword does not correspond to the content. For example, the content registered in the category "musical instrument" does not involve the search keyword "piano".
[0118] The calculation unit 1 is generated, for example, using a model obtained by learning based on a large number of documents. For example, if learning is performed such that the similarity of words used in the same context becomes larger, then even if the meanings are completely different, as long as the frequency of use in the same context is high, the similarity is also high. For example, "spicy" and "sweet" have completely opposite meanings, but because there are many cases where they are used in exactly the same context, such as "like sweet" and "like spicy", the similarity becomes high. Therefore, in the dialogue system 20, the possibility of the above two being associated with the same category is high. However, it is not desirable to include "spicy" in the search keyword of the category "dessert", for example. Therefore, the search keyword "spicy" is removed from the category "dessert".
[0119] In this case, since the error occurred in the calculation unit 1, it is preferable to change the operation of the calculation unit 1. That is, the operation is changed in a manner to reduce the similarity between the category "dessert" and the search keyword "spicy". Since it is not desirable that the similarity between "dessert" and "spicy" is high, which is common content regardless of the domain, the same change is likely to be required in other domains. Therefore, in order to avoid making the same change multiple times, the calculation unit 1 is shared between domains, and the change content is also shared. In this way, the common change content regardless of the domain can be shared, and the same change is not made multiple times.
[0120] If the calculation unit 1 is maintained in common for the domains and the determination unit 3 is maintained for each domain, changes independent of the domain are reflected in the calculation unit 1, and changes dependent on the domain are reflected in the determination unit 3 or the content DB, changes dependent on the domain and changes common to the domain can be realized separately.
[0121] (Variation of the Fourth Embodiment)
[0122] Next, a modification of the fourth embodiment will be described. In the description of the modification, the same description as that of the fourth embodiment will be omitted. In the modification, Example 2 is described in which a change is made to words associated with categories in common for domains and in which the change is made for each domain.
[0123] In the modified example, the calculation unit 1 and the determination unit 3 are held in common for the domains, and the content DB is held for each domain. If changes that are not dependent on the domain are reflected in the calculation unit 1 or the determination unit 3, and changes that are dependent on the domain are reflected in the content DB, changes that are dependent on the domain and changes that are common to the domain can be separated.
[0124] The calculation unit 1 and the determination unit 3 require a large amount of data for learning, and the development costs are high. Therefore, if the data is stored in common for each domain instead of for each domain, only the content DB needs to be created when creating a new domain, and a new domain can be constructed at low cost.
[0125] (Fifth embodiment)
[0126] Next, the fifth embodiment will be described. In the description of the fifth embodiment, the same description as the first embodiment will be omitted. In the first embodiment, the description was made using an example of store search, but in the fifth embodiment, the description is made using an example of a help desk within a company.
[0127] Fig.12 This is a diagram showing an example of display information indicating the association between words and categories according to the fifth embodiment. Fig.12 This is a visualization example of the in-house help desk domain. In store search, the similarity of the meanings of each category is low, and the search keywords are not repeated. Fig.12 As shown, there are many repeated keywords depending on the domain. The analysis device 10 of the fifth embodiment makes it possible to confirm which search keyword is repeated and change it, so that a more preferable dialogue system 20 can be created.
[0128] Finally, an example of the hardware configuration of the analysis device 10 according to the first to fifth embodiments will be described.
[0129] [Hardware configuration example]
[0130] Fig.133 is a diagram showing an example of the hardware configuration of the analysis device 10 according to the first to fifth embodiments. The analysis device 10 according to the first to fifth embodiments includes a control device 301, a main storage device 302, an auxiliary storage device 303, a display device 304, an input device 305, and a communication device 306. The control device 301, the main storage device 302, the auxiliary storage device 303, the display device 304, the input device 305, and the communication device 306 are connected via a bus 310.
[0131] The control device 301 executes the program read from the auxiliary storage device 303 to the main storage device 302. The main storage device 302 is a memory such as ROM (Read Only Memory) and RAM (Random Access Memory). The auxiliary storage device 303 is a HDD (Hard Disk Drive) or a memory card.
[0132] The display device 304 displays display information. The display device 304 is, for example, a liquid crystal display. The input device 305 is an interface for operating the analysis device 10. The input device 305 is, for example, a keyboard and / or a mouse. In the case where the analysis device 10 is a smart device such as a smartphone or a tablet terminal, the display device 304 and the input device 305 are, for example, a touch panel. The communication device 306 is an interface for communicating with other devices.
[0133] The program executed by the analysis device 10 of the first to fifth embodiments is recorded in an installable or executable file on a computer-readable storage medium such as a CD-ROM, memory card, CD-R, and DVD (Digital Versatile Disc), and provided as a computer program product.
[0134] In addition, the program executed by the analysis device 10 of the first to fifth embodiments may be stored in a computer connected to a network such as the Internet and provided by downloading it through the network. In addition, the program executed by the analysis device 10 of the first to fifth embodiments may be provided through a network such as the Internet without downloading it.
[0135] Furthermore, the program of the analysis device 10 according to the first to fifth embodiments may be installed in advance in a ROM or the like and provided.
[0136] The program executed by the analysis device 10 of the first to fifth embodiments is a module structure including the functional blocks that can also be realized by the program among the above-mentioned functional blocks. As for each functional block, as actual hardware, the control device 301 reads the program from the storage medium and executes it, thereby loading the above-mentioned functional blocks into the main storage device 302. That is, the above-mentioned functional blocks are generated on the main storage device 302.
[0137] Furthermore, part or all of the above-mentioned functional blocks may be realized not by software but by hardware such as IC (Integrated Circuit).
[0138] Furthermore, when a plurality of processors are used to implement each function, each processor may implement one of the functions, or may implement two or more of the functions.
[0139] The analysis device 10 of the first to fifth embodiments may be operated in any manner. The analysis device 10 of the first to fifth embodiments may be operated as a cloud system on a network, for example.
[0140] A plurality of embodiments of the present invention have been described, but these embodiments are presented as examples and are not intended to limit the scope of the invention. These new embodiments can be implemented in various other ways, and various omissions, substitutions, and changes can be made without departing from the gist of the invention. These embodiments and their variations are included in the scope and gist of the invention, and are included in the invention described in the claims and the scope equivalent thereto.
[0141] Furthermore, the above-mentioned embodiments can be summarized as the following technical solutions.
[0142] Technical Solution 1
[0143] A parsing device comprising:
[0144] a calculation unit that calculates a first similarity between a first word indicating a category and a second word;
[0145] an applying unit that applies one or more second words having the first similarity greater than a first threshold value to one or more template sentences;
[0146] a determination unit that analyzes a template sentence including the second vocabulary and classifies the second vocabulary into one or more first categories; and
[0147] The display control unit displays the second vocabulary used in the classification of the first category on the display unit for each of the first categories.
[0148] Technical Solution 2
[0149] According to the analytical device described in technical solution 1,
[0150] The device further includes a search unit for searching content based on the first category.
[0151] Technical Solution 3
[0152] According to the analytical device described in technical solution 1,
[0153] The determination unit determines the degree of relevance between the first category and the second vocabulary based on the number of template sentences used when classifying the second vocabulary into the first category,
[0154] The display control unit displays the second word with a degree of relevance assigned thereto.
[0155] Technical Solution 4
[0156] According to the analytical device described in technical solution 1,
[0157] The display control unit displays display information for accepting a change, deletion, or addition of the first category associated with the second vocabulary.
[0158] Technical Solution 5
[0159] According to the analytical device described in technical solution 4,
[0160] The calculation unit changes the first similarity according to change, deletion, or addition of the first category associated with the second vocabulary.
[0161] Technical Solution 6
[0162] According to the analytical device described in technical solution 4,
[0163] The determination unit changes, deletes, or adds the first category into which the second word is classified, according to the change, deletion, or addition of the first category associated with the second word.
[0164] Technical Solution 7
[0165] According to the analytical device described in technical solution 4,
[0166] The system further comprises a content database which associates and stores the content with a second category indicating the classification of the content.
[0167] The display control unit displays display information for accepting a change, deletion, or addition of the first category associated with the second vocabulary,
[0168] The determination unit changes, deletes, or adds the second category according to the change, deletion, or addition of the first category associated with the second word.
[0169] Technical Solution 8
[0170] According to the analytical device described in technical solution 4,
[0171] The display form of the display information is a mapping form in which the relationship between the first category and the second word is represented by the position of the first category and the position of the second word.
[0172] The display control unit receives a change in the first category associated with the second word in response to an input indicating a change in the position of the second word,
[0173] The calculation unit changes the first similarity according to a change in a position of the second vocabulary.
[0174] Technical Solution 9
[0175] According to the analytical device described in technical solution 1,
[0176] The calculation unit receives utterance data, calculates a second similarity between a third word included in the utterance data and the first word, and calculates a first similarity between the first word whose second similarity is greater than a second threshold value and the second word.
[0177] Technical Solution 10
[0178] A parsing method, comprising:
[0179] A step of calculating a first similarity between a first word representing a category and a second word;
[0180] The step of applying one or more second words having the first similarity greater than a first threshold to one or more template sentences;
[0181] A step of parsing the template sentence including the second vocabulary and classifying the second vocabulary into one or more first categories; and
[0182] A step of displaying the second vocabulary used in the classification of the first category on a display unit for each of the first categories.
[0183] Technical Solution 11
[0184] A storage medium stores a program for causing a computer to function as the following units: a calculation unit that calculates a first similarity between a first word indicating a category and a second word;
[0185] an applying unit that applies one or more second words having the first similarity greater than a first threshold value to one or more template sentences;
[0186] a determination unit that analyzes a template sentence including the second vocabulary and classifies the second vocabulary into one or more first categories; and
[0187] The display control unit displays the second vocabulary used in the classification of the first category on the display unit for each of the first categories.
Claims
1. An analysis device comprising: a calculation unit that calculates a first similarity between a first word indicating a category and a second word; an applying unit that applies one or more second words having the first similarity greater than a first threshold value to one or more template sentences; a determination unit that analyzes the template sentence including the second vocabulary and classifies the second vocabulary into one or more first categories; as well as The display control unit displays the second vocabulary used in the classification of the first category on the display unit for each of the first categories.
2. The analysis device according to claim 1, The device further includes a search unit configured to search for content based on the first category.
3. The analysis device according to claim 1, The determination unit determines the degree of relevance between the first category and the second vocabulary based on the number of template sentences used when classifying the second vocabulary into the first category, The display control unit displays the second word with a degree of relevance assigned thereto.
4. The analysis device according to claim 1, The display control unit displays display information for accepting a change, deletion, or addition of the first category associated with the second vocabulary.
5. The analysis device according to claim 4, The calculation unit changes the first similarity according to change, deletion, or addition of the first category associated with the second vocabulary.
6. The analysis device according to claim 4, The determination unit changes, deletes, or adds the first category into which the second word is classified, according to the change, deletion, or addition of the first category associated with the second word.
7. The analysis device according to claim 4, further comprising a content database storing content in association with a second category indicating a classification of the content, The display control unit displays display information for accepting a change, deletion, or addition of the first category associated with the second vocabulary, The determination unit changes, deletes, or adds the second category according to the change, deletion, or addition of the first category associated with the second word.
8. The analysis device according to claim 4, The display form of the display information is a mapping form in which the relationship between the first category and the second word is represented by the position of the first category and the position of the second word. The display control unit receives a change in the first category associated with the second word in response to an input indicating a change in the position of the second word, The calculation unit changes the first similarity according to a change in a position of the second vocabulary.
9. The analysis device according to claim 1, The calculation unit receives utterance data, calculates a second similarity between a third word included in the utterance data and the first word, and calculates a first similarity between the first word whose second similarity is greater than a second threshold value and the second word.
10. A parsing method, comprising: A step of calculating a first similarity between a first word representing a category and a second word; A step of applying one or more second words having the first similarity greater than a first threshold to one or more template sentences; A step of parsing the template sentence containing the second vocabulary and classifying the second vocabulary into one or more first categories; as well as A step of displaying the second vocabulary used in the classification of the first category on a display unit for each of the first categories.
11. A storage medium storing a program for causing a computer to function as the following units: a calculation unit that calculates a first similarity between a first word indicating a category and a second word; an applying unit that applies one or more second words having the first similarity greater than a first threshold value to one or more template sentences; a determination unit that analyzes the template sentence including the second vocabulary and classifies the second vocabulary into one or more first categories; as well as The display control unit displays the second vocabulary used in the classification of the first category on the display unit for each of the first categories.
12. A computer program product, comprising a computer program, wherein the computer program causes a processor to execute the steps of the analysis method according to claim 10.
Citation Information
Patent Citations
Information processing device and information processing method
CN102165437A
Predicate template gathering device, specified phrase pair gathering device and computer program for said devices
CN104137097A
Text classification method, device and equipment and computer readable storage medium
CN110008343A
Interactive device
JP2016085747A
Systems and methods for training a model to determine whether a query with multiple segments comprises multiple distinct commands or a combined command
US20170249297A1
