Voice analysis support device, voice analysis support method and voice analysis support program

Abstract

To provide a voice analysis support device enabling semantic classification of voices according to specific intentions of an analyst in voice analysis such as VOC.SOLUTION: A voice analysis support device for analyzing voice information relating to products or services collected from voice owners, includes acquisition means for acquiring information relating to specific intentions of an analyst regarding a product or a service, acquisition means for acquiring voice information, extraction means for extracting phrases representing voices regarding the product or the service from the voice information, classification means for classifying the extracted phrases into semantic categories represented by the phrases on the basis of the information relating to the specific intentions, and storage means for storing the voice information in association with the phrases and the semantic categories into which the phrase is classified.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present invention relates to a voice analysis support device, a voice analysis support method, and a voice analysis support program. [Background technology]

[0002] VOC (Voice of the Customer) has traditionally been utilized in product development. Product developers analyze VOC collected from customers (clients) via surveys, operator inquiries, complaints, e-commerce site reviews, social media, etc., to help them develop new products that meet customer needs, improve existing products, and develop marketing strategies (PDCA cycle for product development).

[0003] Specifically, when a new product is released, product developers can analyze VOCs received from customers about the new product and analyze what customers think of it. For example, if the product is food, various comments (or opinions, if you will) will be received, such as delicious / not delicious, soft / hard, good / bad texture, quantity too much / too little, appearance good / bad, flavor good / bad, cheap / expensive, and good / bad value for money. By analyzing not only positive comments but also negative ones, product developers can develop better new products, improve existing products, and revise product concepts.

[0004] However, with the recent widespread use of the internet and IT tools, customer feedback has also become big data, and many companies are currently facing challenges with VOC analysis, such as having too much VOC data to organize and classify, not being able to consolidate the VOC data collected from each channel, and not knowing how to analyze and utilize it.

[0005] In recent years, VOC analysis technology has been attracting attention for this reason. It uses natural language processing to analyze huge amounts of text data (VOC data) and extract keywords, and also performs sentiment analysis (positive / negative), allowing for classification and evaluation of text data to make it easier to analyze.

[0006] As a related technology, for example, Patent Document 1 describes a document classification system that classifies text documents into categories by learning classification rules through machine learning without requiring the specification of keywords, etc. for each category, and allows users to easily understand the reasons for the classification results. Also, for example, Patent Document 2 describes a text mining method that can compare emotional trends between multiple documents based on appropriate evaluation of emotional words in the documents with a small amount of calculation. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Application Laid-Open No. 2011-170786 [Patent Document 2] Japanese Patent Application Publication No. 2023-113268 Summary of the Invention [Problem to be solved by the invention]

[0008] Nowadays, in a context where customer needs and preferences for products are becoming increasingly diverse and fragmented, product developers are sometimes developing products that are targeted to specific customer needs and preferences, rather than developing products that are universally popular.

[0009] For example, suppose there is an instant powdered lemon tea product that dissolves well in cold water. The product developer's intention (product concept) was to develop a powdered lemon tea that dissolves well in cold water so that consumers could easily drink iced lemon tea. Next, users (consumers) reported the following VOC cases regarding the instant powdered lemon tea product in question. VOC1: "Dissolves easily in hot water" VOC2: "The lemon flavor is delicious."

[0010] When analyzing such VOC1 and VOC2, if the technology of Patent Document 1 is applied, the text document of VOC1 is classified into a category such as the "usage method (usage characteristics)" of the product by learning classification rules through machine learning, and according to Patent Document 2, VOC1 can be classified into a VOC with an emotional tendency such as "positive." Furthermore, the text document of VOC2 can be classified into a category such as the "taste" of the product, and VOC2 can be classified into a VOC with an emotional tendency such as "positive." Thus, when analyzing VOCs, analysts can specify a category such as "usage method (usage characteristics)" or "taste" or an emotional classification such as "positive," and easily select and analyze VOC1 and VOC2 classified into these categories or emotional classifications from a vast number of VOCs.

[0011] However, according to the invention described in Patent Document 2, for example, regarding VOC1, a customer's comment that "it dissolves well in hot water" is categorized as "positive," even though this comment differs from the product developer's intended usage and characteristics, which are that it dissolves well in cold water. In other words, there is no concept of VOC analysis from the perspective of the product's intention, which is what product developers want to analyze and evaluate. Instead, the system merely categorizes the vast amount of VOC text documents into categories for easier analysis, and performs emotional evaluations based solely on the VOC text documents. This makes it difficult to provide product developers with more advanced analysis results or further insights.

[0012] When product developers develop a product with a specific intention (concept), they can analyze user VOCs from the perspective of that product intention to confirm customer evaluations of that product intention and use them to develop better new products, improve existing products, and revise existing products. Therefore, in the case of VOC1, "dissolves easily in hot water," categorizing it as "positive" may not necessarily be appropriate when considering its relationship to the product intention (dissolves easily in cold water). In this case, the true meaning of VOC1 to the product developer may not be "positive" or "negative," but rather, for example, "miscommunication." In this case, the action that product developers should take based on this realization would be to improve promotions, such as revising product feature descriptions or improving packaging, so that the product concept is accurately conveyed to customers (see Figure 1, for example).

[0013] The present invention has been proposed in view of the above points, and aims to enable semantic classification of voices according to the analyst's specific intentions in voice analysis of VOCs, etc. [Means for solving the problem]

[0014] In order to solve the above problems, the voice analysis support device of the present invention is a voice analysis support device that analyzes voice information regarding a product or service collected from a speaker, and includes an acquisition means for acquiring information regarding the analyst's specific intention regarding the product or service, an acquisition means for acquiring the voice information, an extraction means for extracting phrases indicating the voice regarding the product or service from the voice information, a classification means for classifying the extracted phrases into semantic classifications indicated by the phrases based on the information regarding the specific intention, and a storage means for storing the voice information in association with the phrases and the semantic classifications into which the phrases have been classified. [Effects of the Invention]

[0015] According to one aspect of the embodiment of the present invention, in voice analysis of VOC or the like, it is possible to perform semantic classification of voice according to a specific intention of an analyst. [Brief explanation of the drawings]

[0016] [Figure 1] FIG. 1 is a diagram illustrating the relationship between VOC analysis results and product development strategies according to the present embodiment. [Figure 2] 1 is a diagram illustrating an example of the configuration of a VOC analysis support system 100 according to an embodiment of the present invention. [Figure 3] 1 is a diagram illustrating an example of a functional configuration of a VOC analysis support device 10 according to an embodiment of the present invention. [Figure 4] 10A and 10B are diagrams illustrating an example of product information and information regarding product intention according to the present embodiment. [Figure 5] FIG. 2 is a diagram showing an example of VOC information according to the embodiment; [Figure 6] 10 is a flowchart showing a VOC context classification process according to the present embodiment. [Figure 7A] FIG. 1 is a diagram (part 1) for explaining a context classification process for VOC according to the present embodiment. [Figure 7B] FIG. 10 is a diagram (part 2) for explaining the context classification process for VOC according to the present embodiment. [Figure 7C] FIG. 10 is a diagram (part 3) for explaining the context classification process for VOC according to the present embodiment. [Figure 8] FIG. 10 is a diagram showing an example of a VOC analysis tabulation screen according to the present embodiment. [Figure 9] FIG. 10 is a diagram illustrating an example of a product improvement measure according to the customer attributes of the VOC and the commenter in this embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0017] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described in detail with reference to the drawings. [First embodiment] <Summary> First, a product developer inputs knowledge about a product or service (hereinafter also referred to as "product") for which they wish to perform VOC analysis into a knowledge base, structuring and formatting it in advance as a DB. Knowledge about the product is information about the product intention of the product developer, which is also information about the product analysis perspective that the product developer wants to confirm through VOC analysis. When VOC text data is collected from customers, the VOC analysis support device of this embodiment interprets, classifies, analyzes, aggregates, etc. the VOC in accordance with the knowledge about the product entered into the knowledge base, making it easier to perform VOC analysis from a perspective that is consistent with the product intention of the product developer.

[0018] In this way, by having product developers input and learn knowledge such as product intentions into a knowledge base in advance for VOC analysis, the VOC analysis support device according to this embodiment can interpret and determine the meanings (called contexts) indicated by VOCs in accordance with the product intentions of the product developers, and classify the VOCs into semantic categories (called context categories) indicated by the VOCs for the product developers. This enables product developers to perform VOC analysis in accordance with the analytical perspectives of the products or services they want to focus on during VOC analysis. Furthermore, improving the quality of semantic interpretation of VOCs can be useful for further improving the PDCA cycle of product development.

[0019] Note that "context" generally refers to the context, background, situation, and circumstances surrounding something, and includes vague and broad information such as the context and the culture behind each event. In this embodiment, context refers to the meaning of VOC interpreted in accordance with the product intent in which it is placed. In this embodiment, the meaning of voice of the customer (VOC) is not the literal, single (univocal) meaning of the VOC text; rather, the meaning of the voice of the customer (VOC) that a company wants to analyze can be determined only by considering its relationship to the "product intent" of product developers.

[0020] <System configuration> Fig. 2 is a diagram showing an example of the configuration of a VOC analysis support system 100 according to this embodiment. The VOC analysis support system 100 in Fig. 2 includes a VOC analysis support device 10, a knowledge base 20, and a terminal 30, which are connected via a network (not shown).

[0021] The VOC analysis support device 10 is a server device that supports the analysis of VOCs related to products submitted by customers. The VOC analysis support device 10 performs VOC analysis such as interpretation, judgment, classification, and aggregation of VOCs based on product information stored in a knowledge base 20, knowledge related to products intended by product developers, and VOC information.

[0022] Knowledge base 20 is a knowledge database that accumulates and builds knowledge information about products, such as product master data, information about product intentions of product developers, and VOC information collected from customers. A knowledge base is a database that compiles in-house business experience and know-how in one place, making it easy to search and use.

[0023] Terminal 30 is, for example, a personal computer (PC), smartphone, or tablet terminal, and is a user terminal used, for example, by a product development staff member in a product development department. A predetermined application program and a general-purpose web browser for accessing VOC analysis support device 10 are pre-installed on terminal 30. Product development staff can access VOC analysis support device 10 using terminal 30, obtain analysis results (interpretation, judgment, classification, aggregation, etc.) by VOC analysis support device 10, and display them on the screen. Product development staff analyze VOCs for products submitted by customers to help with the product development PDCA cycle, such as improving existing products to meet customer needs, developing new products, and formulating marketing strategies.

[0024] Needless to say, users of the terminal 30 are not limited to product development personnel, but also include personnel involved in product development, such as product development department, product planning department, product manufacturing department, public relations department, marketing department, etc. Furthermore, although the VOC according to this embodiment is customer feedback on a product, it may also be VOE (Voice of Employee) from employees in the distribution process, including manufacturing, distribution, procurement, and retail (store clerks, etc.), and other personnel involved (delivery personnel, production staff and engineers involved in the manufacturing process).

[0025] (Functional configuration) 3 is a diagram showing an example of the functional configuration of the VOC analysis support device 10 according to this embodiment. The VOC analysis support device 10 has, as its main functional units, a specific intention acquisition unit 101, a VOC acquisition unit 102, a phrase extraction unit 103, a context classification unit 104, a display control unit 105, and a storage unit 109.

[0026] The specific intention acquisition unit 101 has a function of acquiring information regarding the analyst's specific intention regarding a product or service. The VOC acquisition unit 102 has a function of acquiring voice information (VOC). The phrase extraction unit 103 has a function of extracting phrases (words) indicating voice regarding a product or service from the voice information. The context classification unit 104 has a function of classifying phrases into semantic classifications (e.g., context classifications) indicated by the phrases based on the information regarding the specific intention. The display control unit 105 has a function of displaying phrases and semantic classifications into which the phrases have been classified (e.g., context classification processing results) on the terminal 30. The storage unit 109 has a function of storing voice information in the knowledge base 20 in association with phrases and semantic classifications into which the phrases have been classified (e.g., context classification processing results).

[0027] The VOC analysis support device 10 can be implemented using a general-purpose computer. Specifically, the VOC analysis support device 10 includes hardware such as a processing unit (e.g., a CPU), memory, an input / output interface, and a communication interface. The functions of the VOC analysis support device 10 are realized by the processing unit executing processes in accordance with a computer program stored in memory. That is, each functional unit is realized by a computer program executed on hardware resources such as the processing unit and memory of the computer that constitutes the VOC analysis support device 10. The VOC analysis support device 10 may also be referred to as a computing machine for VOC analysis support. These functional units may also be referred to as "means," "module," "unit," or "circuit." The knowledge base 20 may be located in the memory of the VOC analysis support device 10 or in an external storage device on a network. Each functional unit of the VOC analysis support device 10 may not only be implemented by a single server device, but may also be implemented as a system consisting of multiple devices with distributed functions. The computer program may also be stored on a computer-readable storage medium.

[0028] (Knowledge Base) The knowledge base 20 according to this embodiment is a knowledge database in which product information, information relating to product intentions of product developers, VOC information, and the like are accumulated and constructed.

[0029] 4A is a diagram showing an example of product information according to this embodiment. The product information is product master information including, for example, a product code, a product name, a JAN code, a product category, a price, a content amount, a manufacturer, a release date, etc.

[0030] 4(b) is a diagram showing an example of information related to product intentions according to this embodiment. In recent years, customer needs and preferences for products have become increasingly diverse and fragmented. In light of this, product developers use product intentions (e.g., product concepts) to develop products that target specific customer needs and preferences.

[0031] For example, a product developer may want to conduct a VOC analysis to determine whether the instant lemon tea (powder type) product envisioned by the developer has been evaluated by customers as intended in the concept. More specifically, what the product developer wants to analyze in a VOC analysis is customer feedback regarding the fact that the developer has focused particularly on developing a powder that dissolves easily in cold or iced water. Therefore, in such a case, information regarding product intent is structured and formalized in advance as a set of product specifications (product attribute categories) and phrases that indicate product intent, such as "usage / usage properties" - dissolves easily in cold water, "shape" - powder, and "flavor" - crisp and delicious, and then input and saved in the knowledge base 20.

[0032] Note that any method of inputting information into the knowledge base 20 may be used, but for example, a product development staff member can access the VOC analysis support device 10 using a terminal 30 and input information via a predetermined dashboard screen or the like provided by the VOC analysis support device 10. The VOC analysis support device 10 can store, for example, product information, in which the product code of the product and information related to the product intention are associated with each other (FIG. 4).

[0033] Furthermore, input to the knowledge base 20 may be performed by automatically extracting information about product intention from a document (for example, the product development proposal shown in FIG. 4) containing the information. The VOC analysis support device 10 performs natural language processing on the document to extract phrases that indicate product intention and determine the product specifications (product attribute category) to which the phrases belong. The extraction process can utilize libraries using conventional methods, including machine-readable catalogs and corpora, morphological analysis, syntactic analysis, semantic analysis, and contextual analysis. Note that if any information about product intention shown in FIG. 4 that does not correspond to the product intention is automatically extracted, the product development staff may be able to delete it via a predetermined dashboard screen, etc.

[0034] FIG. 5 is a diagram showing an example of VOC information according to the present embodiment. The VOC information is voices from customers collected from channels such as questionnaires, inquiries to operators, complaints, EC site reviews, and SNS. The VOC information according to the present embodiment includes, for example, VOC_ID, product name, product code, channel (type), reception date, name, age, gender, voice content, etc. Note that the VOC information includes information (such as product name and product code) that can identify which product the VOC is for by a method according to the channel for obtaining the VOC. For example, in the case of a web-based channel, the product name and product code are required inputs, or in the case of an inquiry channel to the call center, the operator always confirms the product name and product code.

[0035] Also, the VOC information according to the present embodiment is text data, but it may be data other than text (such as voice and image) by a method according to the channel for obtaining the VOC, as long as it is finally converted into text.

[0036] <Context classification processing of VOC (Semantic interpretation processing of VOC)> FIG. 6 is a flowchart showing the context classification processing of VOC according to the present embodiment. By reading and executing a program that can be realized by the arithmetic processing unit of the VOC analysis support device 10, the following steps (hereinafter referred to as "S") can be realized.

[0037] S1: The VOC analysis support device 10 acquires information regarding the product intention of the target product from the knowledge base 20 using, for example, the product code as a key. The information regarding the product intention according to the present embodiment is a set of, for example, product specifications (product attribute categories) and phrases indicating product intentions in order to express what the product is like (FIG. 4).

[0038] S2: The VOC analysis support device 10 acquires VOC information related to the target product using, for example, the product code as a key from the knowledge base 20. The VOC information according to this embodiment includes, for example, the VOC ID, product name, product code, channel (type), reception date, name, age, gender, and voice content (FIG. 5).

[0039] S3: The VOC analysis support device 10 executes natural language processing to extract phrases (hereinafter referred to as VOC phrases) that express customer opinions from the VOC information (voice content). The extraction process can utilize libraries using conventional methods, including machine-readable catalogs and corpora, morphological analysis, syntactic analysis, semantic analysis, and contextual analysis.

[0040] Generally, a phrase is an expression made up of a series of words or syllables, and is treated as a unit with grammatical meaning, larger than a word but smaller than a sentence. However, in this embodiment, the extraction target is not necessarily limited to a phrase consisting of multiple words, and may be one or more words as long as it expresses the voice of the V customer.

[0041] S4: The VOC analysis support device 10 performs a context classification process based on the information about product intention acquired in S1 and the VOC phrase extracted in S2. More specifically, the device first determines the meaning of the VOC phrase and the product specification to which the VOC phrase belongs. Next, the device determines the context of the VOC phrase (the meaning of the VOC phrase to the product development staff) based on the meaning of the VOC phrase and the phrase indicating the product intention according to the product specification to which the VOC phrase belongs. As a result, the VOC phrase is classified into the determined context category.

[0042] The context classification according to this embodiment is a category for semantically classifying VOC, and examples include "positive (praise)," "negative (complaint)," "neutral," "miscommunication (unintended)," "simple impression," "question," "request," and "other." The meaning of VOC is interpreted and determined in accordance with the product intent, and the VOC is classified into a context classification corresponding to the meaning indicated by the VOC.

[0043] S5: The VOC analysis support device 10 associates the product specifications and context classification for each VOC phrase with each other as the result of the context classification process for each piece of VOC information and stores them in the knowledge base 20. Note that if one VOC contains multiple VOC phrases, each VOC phrase is classified into a context classification corresponding to the meaning indicated by that VOC, and therefore one VOC may be classified into multiple context classifications at the same time. This is also illustrated in Figures 7A to 7C. Next, S3 and S4 will be explained in detail.

[0044] 7A is a diagram (part 1) illustrating the context classification process for VOC according to this embodiment. First, the VOC analysis support device 10 extracts VOC phrases 1002a, such as "It dissolves easily in hot water," "It has a rich taste (and a good aroma)," and "It also has a good aroma," from the voice content 1001a of the VOC information.

[0045] Next, the VOC analysis support device 10 determines to which product specification (product attribute category) 1003a the extracted VOC phrase 1002a belongs. In the example of FIG. 7A, the VOC phrase "dissolves well in hot water" is determined to be a word belonging to, for example, "usage method / usage properties" in the product specifications (product attribute category) based on the meaning of the VOC phrase. Also, the VOC phrase "the taste is rich (and the aroma is) good" is determined to be a word belonging to, for example, "taste" in the product specifications (product attribute category) based on the meaning of the VOC phrase. Also, the VOC phrase "the aroma is good" is determined to be a word belonging to, for example, "aroma" in the product specifications (product attribute category) based on the meaning of the VOC phrase.

[0046] The VOC analysis support device 10 then determines the context 1005a indicated by the VOC phrase (the meaning that the VOC phrase indicates to the product development staff) based on the meaning indicated by each VOC phrase 1002a and the phrase 1004a indicating the product intention corresponding to the product specification 1003a to which the VOC phrase 1002a belongs, and classifies the VOC phrase into the determined context 1005a.

[0047] In the example of Figure 7A, the VOC phrase "dissolves well in hot water" is a word that belongs to the product specification "usage method / usage properties," and in relation to the phrase "dissolves well in cold water" that indicates the product intention, it means the same topic but different things, so it is appropriate to semantically classify it into the context of "miscommunication (misguided intention)" based on the meaning indicated by the VOC phrase.

[0048] When product developers develop a product with a specific intention, they analyze user VOCs from the perspective of that product intention to confirm customer evaluations of that product intention and use this information to develop better new products and improve or revise existing products. The product developers in question aimed to develop a powdered lemon tea that "dissolves well even in cold water." However, the customer feedback did not address whether the product "dissolves well in cold water" in the "usage and characteristics" section, but rather stated that the product "dissolves well in hot water," which is unrelated to the product intention. In other words, the VOC phrase "dissolves well in hot water" could be judged as "positive" based solely on the meaning of the VOC phrase, but when considering its relationship to the phrase "dissolves well in cold water," which indicates the product intention, a "positive" classification is not necessarily appropriate.

[0049] In this case, the insights and actions that product developers can gain from classifying the VOC phrase and the context as "miscommunication" include improving promotion so that the product concept (e.g., "dissolves well even in cold water") is accurately conveyed to customers, such as revising the product feature description or improving the packaging (e.g., C in Figure 1).

[0050] Furthermore, the VOC phrase "The taste is rich and good (and the aroma is also good)" is a word that belongs to the product specification "taste," and in relation to the phrase "crisp and delicious" that indicates the product intention, it means the same topic but different things, so it is appropriate to semantically classify it into the context of "miscommunication (misguided intention)" based on the meaning indicated by the VOC phrase.

[0051] The product developer in question had developed the product with the aim of creating a lemon tea with a crisp and delicious taste, but the customer feedback did not refer to whether the taste was crisp and delicious, but rather to the "rich taste (and aroma) which is unrelated to the product's intention. In other words, the VOC phrase "rich taste (and aroma) is good" could be judged as "positive" based solely on the meaning of the VOC phrase, but when considering its relationship to the phrase "crisp and delicious," which indicates the product's intention, a "positive" classification is not necessarily appropriate.

[0052] In this case, the insights and actions that product developers can gain from classifying the VOC phrase and the context as "miscommunication" include improving promotion so that the product concept (e.g., "crisp and delicious") is accurately conveyed to customers, such as revising the product feature introduction or improving the packaging (e.g., C in Figure 1).

[0053] Here, the technology described in the following non-patent document, for example, can be applied to VOC context classification processing (VOC semantic interpretation processing). This non-patent document describes a metric model (a mathematical model of meaning) that dynamically calculates the semantic and perceptual equivalence, similarity, and relevance between data in accordance with the "situation and context," based on the idea that a single word (e.g., a VOC phrase) does not always have a single meaning, but that the meaning is determined only when the situation and context (information related to product intent) are given. In other words, the semantic and perceptual identity, similarity, and relevance between data are not determined by static relationships, but change dynamically depending on the context and situation.

[0054] Yasushi Kiyoki, "Database System for Measuring Sensibility and Meaning: On the Memory Systems of Humans and Information Systems," "KEIO SFC JOURNAL," Keio University Shonan Fujisawa Society, 2013, Vol. 13, No. 2, pp. 19-26 On the other hand, the VOC phrase "The scent is also good" is a word belonging to the product specification "scent" and does not correspond to the product specification indicating the product intent. In other words, the VOC phrase "The scent is also good" is a reference to the topic or perspective of the product specification "scent" that is unrelated to the product intent, and may be semantically classified into a "positive" context based on the meaning of the VOC phrase. The meaning of the VOC phrase "The scent is also good" is not determined by the VOC phrase alone; as in the above-mentioned non-patent document, the meaning is determined only when the situation and context are given, and it is considered to change dynamically depending on the context and situation. Therefore, it is desirable to apply a metric model (a mathematical model of meaning) that dynamically calculates the semantic and affective equivalence, similarity, and relevance between data in the entire text of the voice content 1001a according to the "situation and context."

[0055] 7B is a diagram (part 2) illustrating the context classification process for VOCs according to this embodiment. First, the VOC analysis support device 10 extracts VOC phrases 1002b, such as "It dissolves easily in water," "The lemon flavor is not tasty," "The volume is small," "The price is a little high," and "It would be helpful if it was reasonably priced," from the voice content 1001b of the VOC information.

[0056] Next, the VOC analysis support device 10 determines which product specification (product attribute category) 1003b the extracted VOC phrase 1002b belongs to. In the example of FIG. 7B , the VOC phrase "dissolves quickly in water" is determined to be a word belonging to, for example, "usage method / usage properties" in the product specifications (product attribute category) based on the meaning of the VOC phrase. The VOC phrase "the lemon flavor is not tasty" is determined to be a word belonging to, for example, "taste" in the product specifications (product attribute category) based on the meaning of the VOC phrase. The VOC phrase "small capacity" is determined to be a word belonging to, for example, "volume" in the product specifications (product attribute category) based on the meaning of the VOC phrase. The VOC phrases "the price is a little high" and "it would be great if it was reasonably priced" are determined to be words belonging to, for example, "price" in the product specifications (product attribute category) based on the meaning of the VOC phrase.

[0057] The VOC analysis support device 10 then determines the context 1005b indicated by the VOC phrase (the meaning that the VOC phrase indicates to the product development staff) based on the meaning indicated by each VOC phrase 1002b and the phrase 1004b indicating the product intention corresponding to the product specification 1003b to which the VOC phrase 1002b belongs, and classifies the VOC phrase into the determined context 1005b.

[0058] In the example of Figure 7B, the VOC phrase "dissolves quickly in water" is a word that belongs to the product specification "usage method / usage properties," and in relation to the phrase "dissolves well even in cold water," which indicates the product intention, it means the same or similar thing on the same topic (positive meaning).Therefore, based on the meaning indicated by the VOC phrase, it is appropriate to semantically classify it into a "positive" context.

[0059] The product developer in question had developed a powdered lemon tea that "dissolves well even in cold water," but customer feedback indicated that it "dissolves quickly in water," which is close to the product's intention. In other words, the VOC phrase "dissolves quickly in water" is appropriately classified as "positive" when considering its relationship with the phrase "dissolves well even in cold water," which indicates the product's intention.

[0060] In this case, the product developer can confirm that the product intent was as intended based on the classification results of the VOC phrase and the "positive" context (e.g., A in Figure 1).

[0061] Furthermore, the VOC phrase "The lemon flavor is not tasty" is a word that belongs to the product specification "taste," and in relation to the phrase "crisp and delicious," which indicates the product intention, it means the same topic but different things, so it is appropriate to semantically classify it into the context of "miscommunication (misguided intent)" based on the meaning indicated by the VOC phrase.

[0062] The product developer in question had developed the product with the aim of creating a lemon tea with a "crisp and delicious" taste, but the customer feedback did not refer to whether the tea was crisp and delicious, but rather to the "lemon flavor not being tasty," which is unrelated to the product's intention. In other words, the VOC phrase "The lemon flavor is not tasty" could be judged as "negative" based solely on the meaning of the VOC phrase, but when considering its relationship to the phrase "crisp and delicious," which indicates the product's intention, a "negative" classification is not necessarily appropriate.

[0063] In this case, the insights and actions that product developers gain from classifying the VOC phrase and the context as "miscommunication" can be utilized in new product development as a separate need, such as improving the taste itself or developing new flavors (e.g., D in Figure 1).

[0064] On the other hand, the VOC phrase "small capacity" is a word that belongs to the product specification "volume" and does not correspond to the product specification that indicates the product intention. In other words, the VOC phrase "small capacity" is mentioned in relation to the topic and perspective of the product specification "volume," which is unrelated to the product intention, and can be semantically classified into a "negative" context based on the meaning indicated by the VOC phrase.

[0065] In addition, the VOC phrase "The price is a little high" is a word that belongs to the product specification "price" and does not correspond to the product specification that indicates the product intention. In other words, the VOC phrase "The price is a little high" is mentioned in relation to the topic or perspective of the product specification "price" that is unrelated to the product intention, and based on the meaning indicated by the VOC phrase, it can be semantically classified into a "negative" context.

[0066] Furthermore, the VOC phrase "It would be helpful if it was reasonably priced" is a word that belongs to the product specification "price" and does not correspond to the product specification that indicates the product intent. In other words, the VOC phrase "It would be helpful if it was reasonably priced" is a reference to the topic or perspective of the product specification "price" that is unrelated to the product intent, and can be semantically classified into the context of "request" based on the meaning indicated by the VOC phrase.

[0067] 7C is a diagram (part 3) illustrating the context classification process for VOC according to this embodiment. First, the VOC analysis support device 10 extracts VOC phrases 1002c, such as "If you don't mix the ice water thoroughly, the powder won't dissolve and some lumps will remain," "The package might be difficult to open," and "I wonder if there's anything we can do about it," from the voice content 1001c of the VOC information.

[0068] Next, the VOC analysis support device 10 determines which product specification (product attribute category) 1003c the extracted VOC phrase 1002c belongs to. In the example of Figure 7C, the VOC phrase "If you don't mix it well with ice water, the powder won't dissolve and some lumps will remain" is determined to be a word belonging to "usage method / usage characteristics" in the product specifications (product attribute category) based on the meaning of the VOC phrase. Also, the VOC phrases "The package might be difficult to open" and "I wonder if they can do something about it" are determined to be words belonging to "packaging" in the product specifications (product attribute category) based on the meaning of the VOC phrase.

[0069] The VOC analysis support device 10 then determines the context 1005c indicated by the VOC phrase (the meaning that the VOC phrase indicates to the product development staff) based on the meaning indicated by each VOC phrase 1002c and the phrase 1004c indicating the product intention corresponding to the product specification 1003c to which the VOC phrase 1002c belongs, and classifies the VOC phrase into the determined context 1005c.

[0070] In the example of Figure 7C, the VOC phrase "If you don't mix it well in ice water, the powder won't dissolve and some lumps will remain" is a word that belongs to the product specification "Usage and Usage Characteristics," and in relation to the phrase "It dissolves well even in cold water," which indicates the product intention, it has the opposite meaning (negative meaning) on ​​the same topic, so it is appropriate to semantically classify it into a "negative" context based on the meaning indicated by the VOC phrase.

[0071] The product developer in question had developed a powdered lemon tea that "dissolves well even in cold water," but customer feedback was that "if you don't mix it well with ice water, the powder doesn't dissolve and some lumps remain," which contradicts the product's intention. In other words, the VOC phrase "if you don't mix it well with ice water, the powder doesn't dissolve and some lumps remain" is appropriately classified as "negative" when considering its relationship with the phrase "dissolves well even in cold water," which indicates the product's intention.

[0072] Also in this case, the product development staff can identify improvements to the product, such as the development of a powder that does not leave lumps even with cold water, from the VOC phrases and the insights and actions obtained from the classification results into the "negative" context (e.g., B in FIG. 1).

[0073] On the other hand, the VOC phrase "Can't we do something?" is a word belonging to the product specification "package" and does not correspond to the product specification indicating the product intention. That is, the VOC phrase "Can't we do something?" is mentioned in relation to the topic / viewpoint of the product specification "package" that has nothing to do with the product intention, and may be classified into the context of "mere impression" (or "request") based on the meaning indicated by the VOC phrase.

[0074] <VOC Analysis Aggregation> FIG. 8 is a diagram showing an example of a VOC analysis aggregation screen according to the present embodiment. The product development staff accesses the VOC analysis support device 10 using the terminal 30, acquires the VOC analysis aggregation information analyzed and aggregated by the VOC analysis support device 10, and displays it on the dashboard screen. The product development staff can analyze the VOC related to the product sent from the customer and analyze the voice of the customer with respect to the product intention (e.g., product concept).

[0075] Specifically, first, the product development staff searches for VOC analysis aggregation information related to the VOC analysis target product using, for example, the product code or product name of the product as a key. In response to the search request for the VOC analysis aggregation information from the terminal 30, the VOC analysis support device 10 can acquire and aggregate the context classification processing results for the product, and display them as the VOC analysis aggregation information shown in FIG. 8, for example.

[0076] Product developers want to conduct a VOC analysis to determine whether or not customers have evaluated the product in accordance with the product intentions envisioned by the product developers. Therefore, when product developers perform a search based on input conditions such as the product code or product name, and product specifications related to the product intentions (for example, "usage method / usage properties," "shape," and "taste"), the VOC collected about the product will be displayed, categorized by product specifications and the contexts that each VOC phrase represents.

[0077] Furthermore, VOC phrases are linked to the original VOC information that originated the VOC phrase, so if product development staff want to check and analyze customer feedback in more detail, they can quickly and easily view the original VOC information (full text) by clicking on the VOC phrase.

[0078] More specifically, a product developer has developed a powdered lemon tea that "dissolves well even in cold water," but according to the results of the context classification process for that product, one of the customer comments is "dissolves well in hot water," which is classified as a "miscommunication" context in "usage method / usage characteristics." After carefully reviewing and analyzing the original VOC information (full text), the product developer can consider improving promotions based on that customer feedback, such as revising the product feature introduction text or improving the packaging, so that the product concept (e.g., "dissolves well even in cold water") is accurately conveyed to customers.

[0079] One of the customer comments is "It dissolves quickly in water," which is classified as a "positive" context in "usage method / usage characteristics." Product developers can confirm that the product has been successful as intended by listening to this customer's feedback. Another customer comment is "If you don't mix it thoroughly in ice water, the powder doesn't dissolve and some lumps remain," which is classified as a "negative" context in "usage method / usage characteristics." Product developers can use this customer's feedback to consider product improvements, such as developing a powder that doesn't dissolve in cold water, realizing that the powder doesn't dissolve properly.

[0080] As described above, according to this embodiment, by predefining information about specific intentions that product developers want to focus on during VOC analysis and inputting and learning it into the knowledge base 20, the VOC analysis support device can classify a huge amount of VOC text data (customer feedback) into the meanings (contexts) that the VOC indicates to the product developers. This enables product developers to perform VOC analysis in line with the perspective of specific intentions related to products or services that they want to focus on during VOC analysis.

[0081] The following embodiments are also mentioned. Generally, a phrase is an expression made up of a series of words or syllables, and is treated as a unit with grammatical meaning, larger than a word but smaller than a sentence. However, the phrase in this embodiment is not necessarily limited to a combination of words that make sense; it can be a single word or a phrase as long as it expresses the customer's voice. In other words, a phrase can make sense either as a single word or as a combination of words, and is a semantic unit determined by the result of determining its meaning.

[0082] The VOC phrase extraction process in S3 analyzes VOC sentences > segmented sentences > phrases > words, taking into account the initial information provided by the language, such as nouns, verbs, adjectives, and adverbs, to extract meaningful VOC phrases. Furthermore, when a VOC phrase is combined, its meaning changes depending on the center of gravity of the other words, the situation, and the context. It is only after these factors are taken into account that the meaning is determined. For example, in a VOC about instant lemon tea, the true meaning of the VOC phrases "good value for money" and "poor value for money" cannot be determined simply from the price alone. The true context of the VOC is determined by considering various other circumstances and contexts, such as the full VOC text, other phrases, and information about the product's intent. Therefore, depending on the situation and context, the true context of the VOC may be the exact opposite or a miscommunication (misunderstanding of the intended meaning).

[0083] For example, when there is a huge amount of VOC, some of the comments may overlap. On the VOC analysis summary screen, VOCs with the same meaning may be aggregated and the VOC phrases may be displayed together. In this case, the number of VOCs may be displayed together with the VOC phrase, for example, "sweet" (50 entries). Product developers can understand the importance of a VOC phrase based on the number of VOCs that indicate a large or small number of comments.

[0084] Product specifications (product attribute categories) vary depending on the product itself being analyzed for VOCs. Examples include other quality-related factors, product design and labeling, containers and packaging and materials, content volume, size, and quantity, design, manufacturing method and process, usage and storage methods, return and exchange procedures, communication, product lineup and inventory status, etc. For example, if the product is an ingredient or seasoning, and the product developer defines the product intent they want to focus on in their VOC analysis as whether it is suitable for cooking or recipes, product specifications could include the genre of food (Japanese, Western, Chinese, etc.) or the specific name of the dish.

[0085] Context classification is not limited to the examples in the above-mentioned embodiments, but varies depending on the perspective of specific intentions that product developers want to focus on during VOC analysis, i.e., the information needs that companies want to know through VOC analysis. In other words, product developers can perform VOC analysis that meets the various information needs of companies depending on how they define context to classify information related to specific intentions and products, etc.

[0086] In VOC analysis, for example, product evaluations may change between immediately after a product is released and after a certain period of time has passed since the product is released, due to factors such as customer understanding of the product concept becoming more widespread. In response to a request from the terminal 30, the VOC analysis support device 10 may analyze and aggregate VOCs over a predetermined period (for example, monthly), and display the VOC analysis and aggregation information for each predetermined period in chronological order on the screen.

[0087] By utilizing the VOC analysis support system 100 according to this embodiment, not only for new product launches but also before and after product renewal, it is possible to analyze changes in VOCs before and after product renewal and whether the intention behind the renewal is close or far from being true.

[0088] The VOC analysis support system 100 (VOC analysis support device 10) of this embodiment is not limited to utilizing VOC in product and service development, but can also be applied to utilizing the opinions received from VOE of employees (store clerks, etc.) and other related parties (delivery staff, manufacturing staff and engineers involved in the manufacturing process, etc.) in the distribution process, including manufacturing, distribution, procurement, and retail.

[0089] Furthermore, by separating voices into VOC and VOE, and comparing and aggregating different voices from different sources, it becomes possible to identify needs, issues, problems, etc. that would be difficult to identify from a single voice source alone.

[0090] [Second embodiment] As mentioned above, product developers can use VOC analysis to develop new products that meet customer needs, improve existing products, and plan marketing strategies. By gaining a high-resolution understanding of who is expressing the VOC, product developers can more accurately consider how to improve products in the future.

[0091] FIG. 9 is a diagram illustrating an example of product improvement measures based on VOC and the customer attributes of the commenter according to this embodiment. For example, if VOCs related to the context classification "taste": salty (negative) and VOCs related to "taste": bland (negative) are received for the product "pre-packed seasoned meatballs," an important insight that a product developer can gain from these VOCs is product improvement measures for the "taste" aspect, such as simply reducing the amount of salt or making the seasoning stronger. However, by understanding "who" made the comments, product developers can consider how to improve the product in the future from a broader perspective based on seemingly contradictory VOCs.

[0092] For example, regarding the product "pre-packaged seasoned meatballs," if you understand that the VOC commenting "taste" as "salty" (negative) came from a "mother with a kindergarten child," it would be possible to consider more in-depth product improvement measures, such as considering a recipe with less salt, taking into account the growth process and development of a child's sense of taste, as well as changing the packaging design to a cute and pop design using characters and adding the phrase "low in salt."

[0093] For example, regarding the product "ready-to-eat seasoned meatballs," if you understand that the VOC commenting that "flavor" is weak (negative) was made by an "elderly person," it would be possible to consider more in-depth product improvement measures, such as considering a recipe that adds umami flavor, taking into account the declining sense of taste and excessive salt intake among the elderly, and adding the words "can be eaten every day" to the packaging with large, easy-to-read letters that give a healthy impression.

[0094] Generally, VOC is collected from sources such as surveys, inquiries to operators, e-commerce site reviews, and social media, but in the case of VOC collected from surveys (especially questionnaire forms) and inquiries to operators, the customer attributes of the VOC voices are often unknown or insufficient. On the other hand, when VOC is linked to customer IDs and customer attribute information (user profiles, etc.), such as VOC collected from e-commerce site reviews and social media, it is possible to identify the customer attributes of the voices in that VOC.

[0095] Examples of customer attribute information that can be particularly useful in considering product improvement measures include age, gender, family structure, address (area of ​​residence), and occupation. Customer attributes of voices in VOC can also be classified into clusters. Examples of cluster classification information include young singles, DINKS, families with children, families with children (dual-income families), two-generation families, elderly singles, and elderly couples.

[0096] On the other hand, there are cases where customer attribute information (user profile, etc.) does not contain enough information to contribute to the consideration of product improvement measures, for example, family structure or occupation is left blank or not registered. In such cases, by applying patents such as Patent No. 6923890 and Patent No. 6971449 owned by the present applicant, if the customer is a member of an e-commerce site selling products, it is possible to estimate the customer's family structure, occupation, etc. based on the customer's activity data, such as a purchase history database or transaction data, and to supplement the blank or not registered customer attribute information.

[0097] As shown in Figure 8, VOC phrases are linked to the original VOC information from which the VOC phrase originated, and by clicking on the VOC phrase, it is possible to quickly and easily check the original VOC information (full text), as well as the customer attribute information (user profile, etc.) and cluster classification information of the person who spoke the VOC.

[0098] Although the present invention has been described with reference to specific examples according to the preferred embodiments of the present invention, it is apparent that various modifications and changes can be made to these examples without departing from the broad spirit and scope of the present invention as defined in the appended claims. In other words, the details of the examples and the accompanying drawings should not be construed as limiting the present invention.

[0099] (Appendix 1) A voice analysis support device (e.g., a VOC analysis support device) that analyzes voice information (e.g., VOC) about a product or service collected from a voice source (e.g., a customer), An acquisition means (e.g., S1 in FIG. 6) for acquiring information regarding a specific intention of an analyst (e.g., a product development manager) for the product or service (e.g., a phrase indicating the product intention, or a set of product specifications and phrases indicating the product intention); An acquisition means (for example, S2 in FIG. 6) for acquiring the voice information (for example, VOC text data); An extraction means (for example, S3 in FIG. 6) for extracting phrases (for example, VOC phrases) indicating opinions about the product or service from the voice information; A classification means (e.g., S4 in FIG. 6) that classifies the extracted phrase into a semantic classification (e.g., a context classification) indicated by the phrase based on the information regarding the specific intention; a storage means (e.g., S6 in FIG. 6) for storing the voice information in association with the phrase and the semantic classification into which the phrase has been classified (e.g., a result of a context classification process); A voice analysis support device comprising:

[0100] (Appendix 2) a display control means for displaying the phrases and the semantic classifications into which the phrases have been classified (e.g., context classification processing results) on the terminal of the analyst; 2. The voice analysis support device according to claim 1, comprising:

[0101] (Appendix 3) The classification means determining a semantic classification category into which the phrase is classified among predetermined semantic classification categories (e.g., "positive (praise)," "negative (complaint)," "neutral," "miscommunication (misintentional)," "simple impression," "question," "request," and "other") based on the information about the specific intention; 2. The voice analysis support device according to claim 1,

[0102] (Appendix 4) the predetermined semantic classification categories include a first semantic classification corresponding to the specific intention (e.g., “positive (praise)” and “negative (complaint)”) and a second semantic classification not corresponding to the specific intention (e.g., “miscommunication (off-purpose)”), The classification means If the meaning of the phrase corresponds to the specific intent, classify the phrase into the first semantic classification (e.g., context classification “positive (praise)” or “negative (complaint)”); classifying the phrase into the second semantic classification (e.g., the context classification "miscommunication") if the meaning of the phrase does not correspond to the specified intent; 4. The voice analysis support device according to claim 3,

[0103] (Appendix 5) and an acquisition means for acquiring attribute information of the speaker (for example, "a mother with a kindergarten child" or "an elderly person who is not aware of the decline in taste"); The display control means further displaying attribute information of the voice owner on the terminal of the analyst; 3. The voice analysis support device according to claim 2, [Explanation of symbols]

[0104] 10 VOC analysis support equipment 20 Knowledge Base 30 devices 100 VOC analysis support system 101 VOC acquisition department 102 Phrase Extraction Unit 103 Context Acquisition Unit 104 Context Classification Unit 105 Display control unit 109 Storage section

Claims

1. A voice analysis support device that analyzes voice information related to a product or service collected from a voice source, An acquisition means for acquiring information regarding the analyst's specific intention regarding the product or service; an acquisition means for acquiring the voice information; an extraction means for extracting phrases indicating comments about the product or service from the comment information; a classification means for classifying the extracted phrase into a semantic classification indicated by the phrase based on the information regarding the specific intention; a storage means for storing the voice information in association with the phrase and the semantic classification into which the phrase is classified; A voice analysis support device comprising:

2. a display control means for displaying the phrases and the semantic classifications into which the phrases are classified on the terminal of the analyst; 2. The voice analysis support device according to claim 1, further comprising:

3. The classification means determining a semantic classification category into which the phrase is classified from among predetermined semantic classification categories based on the information about the specific intention; 2. The voice analysis support device according to claim 1, wherein:

4. the predetermined semantic classification category includes a first semantic classification corresponding to the specific intention and a second semantic classification not corresponding to the specific intention, The classification means If the meaning of the phrase corresponds to the specific intent, classify the phrase into the first semantic classification; classifying the phrase into the second semantic classification if the meaning of the phrase does not correspond to the specific intent; 4. The voice analysis support device according to claim 3, wherein:

5. and acquiring means for acquiring attribute information of the voice owner, The display control means further displaying attribute information of the voice owner on the terminal of the analyst; 3. The voice analysis support device according to claim 2, wherein:

6. A voice analysis support method for analyzing voice information related to a product or service collected from a voice source, comprising: The voice analysis support device An acquisition means for acquiring information regarding the analyst's specific intention regarding the product or service; an acquisition step of acquiring the voice information; an extraction step of extracting phrases indicating comments about the product or service from the comment information; a classification step of classifying the extracted phrases into semantic classifications indicated by the phrases based on the information about the specific intention; a storage step of storing the voice information in association with the phrase and the semantic classification into which the phrase is classified; A voice analysis support method characterized by executing the above.

7. Computer, An acquisition means for acquiring information regarding the analyst's specific intention regarding the product or service; An acquisition means for acquiring voice information about the product or service collected from voice owners; an extraction means for extracting phrases indicating comments about the product or service from the comment information; a classification means for classifying the extracted phrase into a semantic classification indicated by the phrase based on the information regarding the specific intention; a storage means for storing the voice information in association with the phrase and the semantic classification into which the phrase is classified; A voice analysis support program to make it function as a voice analysis system.

Images