Intelligent after-sales customer service method, system, storage medium and intelligent terminal

Abstract

The present application relates to an intelligent after-sales customer service method, system, storage medium and intelligent terminal, and relates to the field of after-sales service, which includes obtaining question content; disassembling question keywords; distinguishing keyword group categories; determining the modulus database number; searching for the modulus; forming an overall question vector; performing a two-norm similarity calculation based on the overall question vector and all preset standard vectors to obtain a suitable standard vector; searching for corresponding suitable reply content from a preset answer database based on the suitable standard vector; and outputting the suitable reply content. The present application has the function of obtaining the closest question stored in the database through two-norm similarity calculation, and outputting the answer to the corresponding question, so that all questions can be converted into standard questions, effectively understanding customer problems, improving the ability to handle complex problems, and improving the intelligent effect of after-sales customer service.

Description

Technical Field

[0001] The present application relates to the field of marketing after-sales service, and in particular to an intelligent after-sales customer service method, system, storage medium and intelligent terminal. Background Art

[0002] As industrial manufacturing transforms towards digitalization and intelligence, the demand for intelligent product sales and after-sales services is increasing. Intelligent AI services have indeed been introduced into the after-sales service of industrial products. These intelligent AI services are designed to improve the efficiency and quality of after-sales services, help companies reduce operating costs, and provide customers with a better service experience.

[0003] Traditional after-sales AI services use a preset FAQ (Frequently Asked Questions) library, and intelligent customer service robots can quickly respond to customers' simple inquiries, that is, the content of the questions is fixed, and the answers to the questions are also fixed. When it comes to complex questions or questions that are not in the FAQ library, customers are advised to transfer to manual services for easier answers.

[0004] The following problems exist in the existing technology: once the user changes the word order or grammar, or uses different expressions to express the same content, the traditional after-sales AI service cannot effectively understand the user's needs and has difficulty in handling complex problems. There is still room for improvement. Summary of the invention

[0005] In order to improve the problem that when the same content is expressed in different ways, user needs cannot be effectively understood and complex issues cannot be handled, the present application provides an intelligent after-sales customer service method, system, storage medium and intelligent terminal.

[0006] In the first aspect, the present application provides an intelligent after-sales customer service method, which adopts the following technical solution:

[0007] An intelligent after-sales customer service method, comprising:

[0008] Get the input question content;

[0009] Decompose the question content based on the reference words in the preset industry knowledge base to obtain the question keywords;

[0010] Identify the question keywords based on the action words and state words in the preset specific identification word library to determine the keyword group word category corresponding to the question keywords;

[0011] Determine the corresponding module database number based on the keyword group category;

[0012] Based on the question keywords, the corresponding modulus is found from the database corresponding to the modulus database number;

[0013] The modulus of the question keywords forms the overall question vector;

[0014] Based on the overall problem vector and all preset standard vectors, a two-norm similarity calculation is performed to obtain a standard vector with the greatest similarity, and the standard vector is defined as a suitable standard vector;

[0015] Find the corresponding appropriate reply content from the preset answer database based on the appropriate standard vector;

[0016] Output the appropriate response content.

[0017] By adopting the above technical solution, by determining the keyword category and the location of the keyword (subject, predicate, object), and then determining the corresponding vector, the closest question stored in the database is obtained by calculating the bi-norm similarity, and the answer to the corresponding question is output, so that all questions can be converted into standard questions, effectively understanding customer problems, improving the ability to handle complex problems, and improving the intelligence of after-sales customer service.

[0018] Optionally, a method for obtaining a suitable standard vector based on the overall problem vector is also included, the method comprising:

[0019] Determine the missing word category based on matching the keyword word category and the preset required category;

[0020] When the missing word category does not exist, a suitable standard vector is obtained based on the overall question vector;

[0021] When the missing word group category exists, searching the corresponding subcategory quantity and subcategory content from a preset subcategory database based on the missing word group category;

[0022] When the number of subcategories is equal to 1, a suitable standard vector is obtained based on the overall problem vector;

[0023] When the number of subcategories is greater than 1, the corresponding inquiry words are searched from the preset inquiry database based on the missing word group category;

[0024] Generate and output adapted inquiry dialogue based on question content, subcategory content and inquiry dialogue.

[0025] By adopting the above technical solution, if a keyword is missing in the middle and it is impossible to determine what is missing based on the existing context, it is necessary to ask the customer, which improves the clarity of complex issues.

[0026] Optionally, when the number of subcategories is greater than 1, the method for generating and outputting an adaptive inquiry speech based on the question content and the inquiry speech also includes:

[0027] Get historical question content when the number of subcategories is greater than 1;

[0028] When there is historical problem content, determining a keyword grouping category based on the historical problem content, and defining the keyword grouping category as a historical keyword grouping category;

[0029] Matching based on historical keyword grouping categories and missing keyword grouping categories;

[0030] When there is a historical keyword grouping category that matches the missing grouping category, the question keyword corresponding to the historical keyword grouping category is output as the question keyword corresponding to the missing grouping category, and the inquiry words are not searched;

[0031] When there is no historical keyword group category that matches the missing group category, an adaptive inquiry dialogue is generated based on the question content, subcategory content and inquiry dialogue and output.

[0032] By adopting the above technical solution, it is also possible to determine the missing parts in the current content based on historical issues, so that complete technical issues can be generated without having to repeat all the content each time during the conversation. Context tracking and multi-round conversation management technology can be used to achieve continuous and accurate customer service.

[0033] Optionally, when there is no historical keyword word group category matching the missing word group category, a method for generating and outputting an adaptive inquiry speech based on the question content, the subcategory content and the inquiry speech includes:

[0034] Based on the keywords and keyword group categories corresponding to the historical keyword group categories, the subcategory contents are screened in turn to obtain the number of screened subcategories, and the number of subcategories is defined as the number of screened subcategories;

[0035] When the number of subcategories to be screened is still greater than 1, the subcategory contents are screened in sequence based on the keywords corresponding to the historical keyword grouping categories and the keywords corresponding to the keyword grouping categories;

[0036] When the number of subcategories is still greater than 1 after all the keywords corresponding to the historical keyword grouping categories and the keywords corresponding to the keyword grouping categories have been screened, the subcategory content corresponding to the current number of subcategories is defined as the screened subcategory content;

[0037] Generate and output the adapted inquiry script based on the question content, filtered subcategory content and inquiry script;

[0038] After the keywords corresponding to several historical keyword word grouping categories or the keywords corresponding to the keyword word grouping categories are screened, when the number of screened subcategories is equal to 1, the subcategory content at this time is defined as the only subcategory content;

[0039] The unique subcategory content is output as the question keyword corresponding to the missing word group category, and the inquiry language is not searched.

[0040] By adopting the above technical solution, possible situations of missing parts can also be screened under the setting conditions of other keywords, so that accurate customer service can be achieved without sending inquiry scripts, the degree of human participation is reduced, and the intelligence of after-sales service is improved.

[0041] Optionally, when the number of subcategories is greater than 1, the method for generating and outputting an adaptive inquiry speech based on the question content and the inquiry speech also includes:

[0042] When the missing word group category and the preset picture analyzable category are the same, the input question picture is obtained;

[0043] Matching is performed based on the problem image, the missing word group category and the preset analyzable features to obtain the actual analyzable features;

[0044] Based on the actual analyzable features, corresponding analyzable feature keywords are searched from a preset keyword database respectively;

[0045] Output the analyzable feature keywords as the question keywords corresponding to the missing word group category, without searching for the inquiry words;

[0046] When the missing word group category and the image analyzable category are different, an adaptive inquiry dialogue is generated based on the question content, subcategory content and inquiry dialogue and output.

[0047] By adopting the above technical solution, if there is a picture, see if it can be analyzed from the picture. Since the content that can be analyzed from the picture is limited, when it is a category that can be analyzed from the picture, the corresponding content can be directly analyzed from the picture as the missing content, which improves the flexibility of after-sales service.

[0048] Optionally, further after-sales customer service method is also included when obtaining the input problem picture, the method comprising:

[0049] Matching the problem image with the preset analyzable features to obtain all the analyzable features in the image;

[0050] Based on the analyzable features in the graph, corresponding analyzable feature keywords in the graph are searched from the keyword database respectively;

[0051] Based on the action words and state words in the specific identification word library, the question keywords are identified to determine the keyword group category in the graph corresponding to the analyzable characteristic keywords in the graph;

[0052] Based on the analyzable characteristic keywords in the graph, the corresponding modulus in the graph is searched from the database corresponding to the modulus database number;

[0053] Determining a modulus relationship between the modulus within the graph and the modulus between the moduli based on the keyword group word category within the graph and the keyword group word category;

[0054] When the modulus relationship is a preset similarity relationship or the same relationship, no operation is performed;

[0055] When the modulus relationship is a preset complete distinction relationship, the corresponding in-graph keyword group category is defined as an abnormal keyword group category, and the in-graph analyzable feature corresponding to the abnormal keyword group category is defined as an abnormal feature;

[0056] The area in the problem image corresponding to the abnormal features is marked and output with the preset warning signal.

[0057] By adopting the above technical solution, when there is a picture, it can also be checked whether the content analyzed previously is normal. If the content input by the customer and the content in the picture are inconsistent, an alarm signal is output, thereby improving the accuracy of data verification.

[0058] Optionally, an after-sales customer service method is also included when the similarity corresponding to the question keyword or the appropriate standard vector cannot be obtained and is less than a preset unconfirmable critical value, the method comprising:

[0059] Disassemble the question content based on the reference words in the preset industry knowledge base to obtain some question keywords;

[0060] Based on some of the question keywords, the corresponding label human service personnel are found from the preset label database;

[0061] Send questions to label service personnel and receive manual replies;

[0062] Output the manual reply content, and obtain the corresponding question keywords, keyword group categories, modulus database numbers and moduli based on the manual reply content and the preset disassembly inquiry operation, and record them in the corresponding specific identification word library and modulus database number;

[0063] The manual responses are stored in the answer database.

[0064] By adopting the above technical solution, if it is impossible to rely on the intelligent customer service to answer, the call will be transferred to the manual customer service, and then the content of the handling of the manual customer service after the transfer will be used as reference material for absorption and learning, thereby improving the self-learning ability of the intelligent customer service.

[0065] In the second aspect, the present application provides an intelligent after-sales customer service system, which adopts the following technical solutions:

[0066] An intelligent after-sales customer service system, comprising:

[0067] The acquisition module is used to obtain question content, question images and manual response content;

[0068] A memory, used to store a program of a control method of any of the above-mentioned intelligent after-sales customer service methods;

[0069] The program in the processor and the memory can be loaded and executed by the processor to implement the control method of any of the above-mentioned intelligent after-sales customer service methods.

[0070] By adopting the above technical solution, by determining the keyword category and the location of the keyword (subject, predicate, object), and then determining the corresponding vector, the closest question stored in the database is obtained by calculating the bi-norm similarity, and the answer to the corresponding question is output, so that all questions can be converted into standard questions, effectively understanding customer problems, improving the ability to handle complex problems, and improving the intelligence of after-sales customer service.

[0071] In the third aspect, the present application provides a smart terminal, which adopts the following technical solution:

[0072] The intelligent terminal includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute any of the above-mentioned intelligent after-sales customer service methods.

[0073] By adopting the above technical solution, by determining the keyword category and the location of the keyword (subject, predicate, object), and then determining the corresponding vector, the closest question stored in the database is obtained by calculating the bi-norm similarity, and the answer to the corresponding question is output, so that all questions can be converted into standard questions, effectively understanding customer problems, improving the ability to handle complex problems, and improving the intelligence of after-sales customer service.

[0074] Fourthly, the present application provides a computer storage medium that can store corresponding programs and has the characteristics of fast interaction with memory big data.

[0075] Computer readable storage medium, using the following technical solution:

[0076] A computer-readable storage medium stores a computer program that can be loaded by a processor and execute any of the above-mentioned intelligent after-sales customer service methods.

[0077] By adopting the above technical solution, by determining the keyword category and the location of the keyword (subject, predicate, object), and then determining the corresponding vector, the closest question stored in the database is obtained by calculating the bi-norm similarity, and the answer to the corresponding question is output, so that all questions can be converted into standard questions, effectively understanding customer problems, improving the ability to handle complex problems, and improving the intelligence of after-sales customer service.

[0078] In summary, the present application includes at least the following beneficial technical effects:

[0079] By calculating the similarity of the two norms, we can get the closest question stored in the database, and output the answer to the corresponding question, so that all questions can be converted into standard questions, effectively understand the customer's questions, improve the ability to handle complex problems, and improve the intelligence of after-sales customer service;

[0080] Determine the missing parts of the current content based on historical questions, so that complete technical questions can be generated without having to repeat all the content every time during the conversation. Use context tracking and multi-round conversation management technology to achieve continuous and accurate customer service;

[0081] When it is a category that can be analyzed from the picture, the corresponding content can be directly analyzed from the picture as the missing content, which improves the flexibility of after-sales service. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is a flow chart of an intelligent after-sales customer service method in an embodiment of the present application.

[0083] Figure 2 It is a flowchart of a method for determining whether to obtain a suitable standard vector based on the overall problem vector in an embodiment of the present application.

[0084] Figure 3 It is a flowchart of a method for generating and outputting adaptive inquiry speech based on question content and inquiry speech when the number of subcategories is greater than 1 in an embodiment of the present application.

[0085] Figure 4 It is a flowchart of a method for generating and outputting adaptive inquiry words based on question content, subcategory content and inquiry words when there is no historical keyword word category matching the word category or when the word category is missing in an embodiment of the present application.

[0086] Figure 5 It is a flowchart of a further method for generating and outputting adaptive inquiry speech based on question content and inquiry speech when the number of subcategories is greater than 1 in an embodiment of the present application.

[0087] Figure 6 It is a flowchart of a further after-sales customer service method when obtaining an input question picture in an embodiment of the present application.

[0088] Figure 7 It is a flowchart of the after-sales customer service method in an embodiment of the present application when the question keyword cannot be obtained or the similarity corresponding to the suitable standard vector is less than the preset unconfirmable critical value.

[0089] Figure 8It is a system module diagram of an intelligent after-sales customer service method in an embodiment of the present application. DETAILED DESCRIPTION

[0090] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1-Figure 8 It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0091] The present application embodiment discloses an intelligent after-sales customer service method. Figure 1 , an intelligent after-sales customer service method includes:

[0092] Step 100: Obtain input question content.

[0093] The question content refers to the question content about the purchased product. The customer who purchased the product inputs the question into the after-sales service interface of the corresponding app or webpage, and then converts it into content that can be recognized by the system.

[0094] Step 101: Decompose the question content based on reference words in a preset industry knowledge base to obtain question keywords.

[0095] The industry knowledge base is a database of industry knowledge content in the corresponding product, which is equivalent to a space for storing data and contains reference words. Reference words are deformed words or dialects that may be derived from a certain official word, for example: bolts can also be called screws. Question keywords are key words in the question content.

[0096] The decomposition method is to match the words in the question content one by one, and when the match is successful, the corresponding question keywords are obtained.

[0097] Step 102: Identify the question keyword based on the action words and state words in the preset specific identification word library to determine the keyword group category corresponding to the question keyword.

[0098] The specific identification word library is a database that stores some specific words, such as action words and state words. It is manually created. Action words are words that represent an action, such as: operate, execute, solve. State words are words that represent a certain state, such as: no.

[0099] The keyword group category is the category of the word corresponding to the question keyword in a sentence, for example: subject, predicate, and object. And further subdivided into simple subject, infinitive phrase as subject, gerund phrase as subject, clause as subject, simple predicate, modal verb + main verb, linking verb + predicate, auxiliary verb + main verb, direct object, indirect object, compound object, and double object, etc.

[0100] The identification method is to match the combination of direct text + text.

[0101] Step 103: Determine the corresponding modulus database number based on the keyword group category.

[0102] The modulus database number is the number of the database that can express the corresponding modulus of a certain word in this keyword group category. The purpose of this number here is that there are many overall modulus databases, and the number is to facilitate the search for the corresponding modulus.

[0103] Step 104: Based on the question keyword, the corresponding modulus is searched from the database corresponding to the modulus database number.

[0104] The modulus is a numerical value in the direction corresponding to the question keyword. The purpose is to express the corresponding keyword so that it becomes a vectorized vocabulary. When the moduli corresponding to the question keywords corresponding to the same keyword group category are closer, the overall. The mapping relationship between the question keywords and the modulus is stored in the database. The modulus corresponding to each question keyword in the same keyword group category is set by the staff in this field.

[0105] Step 105: Form an overall question vector based on the modulus of the question keywords.

[0106] The overall question vector is the vector corresponding to the entire question content. The questions in the imported content data are segmented; the questions in the content data are vectorized based on the results of the segmentation, and the segmentation results and importance are also included in the vectorized representation based on RF-IDF.

[0107] Step 106: Perform a two-norm similarity calculation based on the overall problem vector and all preset standard vectors to obtain a standard vector with the greatest similarity, and define the standard vector as a suitable standard vector.

[0108] The standard vector is the content after the standard content of all problems is vectorized.

[0109] The bi-norm (also called the Euclidean norm) is a metric in vector space that is used to measure the length or size of a vector. When calculating the similarity between two vectors, we usually do not use the bi-norm itself directly, but use a related metric - such as the Euclidean distance, which is defined based on the bi-norm. For example, the vector representation of problem 1 is A(x1,x2,…,xn), and the vector representation of problem 2 is B(y1,y2,…,yn). The distance between AB, |AB|=sqrt((x1-y1)^2+(x2-y2)^2+…+(xn-yn)^2). The closer the distance, the higher the similarity.

[0110] Step 107: Find the corresponding appropriate reply content from the preset answer database based on the appropriate standard vector.

[0111] The appropriate reply content is the reply content that can answer the question content corresponding to the appropriate standard vector. The database stores the mapping relationship between the appropriate standard vector and the appropriate reply content. When the staff in this field obtains the question content corresponding to the appropriate standard vector, they reply based on experience and the actual content of the product as the appropriate reply content, and store it in the database after forming a mapping relationship with the appropriate standard vector. When the system receives the corresponding appropriate standard vector, it automatically finds the corresponding appropriate reply content from the database and outputs it.

[0112] Step 108: Output the appropriate reply content.

[0113] Reference Figure 2 , and also includes a method for obtaining a suitable standard vector based on the overall problem vector, the method comprising:

[0114] Step 200: Determine the missing word group category based on matching the keyword word group category and the preset required category.

[0115] The necessary category is a category that must exist, for example: subject, predicate, object. The missing word category is a word category that is missing in addition to the keyword word category, so that the two together form the necessary category. The determination method is that when the two are matched successfully, they do not fall into the missing word category, and when there is a category in the necessary category that is different from the keyword word category, then this category is the missing word category.

[0116] Step 201: When the missing word group category does not exist, obtain a suitable standard vector based on the overall question vector.

[0117] When the missing word group category does not exist, it indicates that the dimension of the question vector is correct, and then the two-norm similarity calculation method can be used for calculation, and then step 106 can be directly used to obtain a suitable standard vector through the overall question vector.

[0118] Step 202: When a missing word group category exists, searching a preset subcategory database for the corresponding subcategory quantity and subcategory content based on the missing word group category.

[0119] The subcategory content is the content of the subcategory of the sub-field corresponding to the missing word group category, for example: the missing word group category is the object, such as: product, then the subcategory content is the type, specification, quantity, etc. of the product. The number of subcategories is the number corresponding to the subcategory content. For example, if the subcategory is the type, specification or quantity of the product, the number of subcategories is 3. Here, when the subcategory corresponding to the missing word group category does not exist, the subcategory content is the specific content corresponding to the missing word group category, and the number of subcategories is the number of corresponding specific content.

[0120] The database stores the mapping relationship between the missing word group category, the number of subcategories, and the subcategory content. The staff in this field confirms the subcategory content of each missing word group category based on their own experience and the products produced by the company, and then counts the number of corresponding subcategory content and records it in the database as the number of subcategories. When the system receives the corresponding missing word group category, it automatically finds the corresponding number of subcategories and subcategory content from the database and outputs them.

[0121] Step 203: When the number of subcategories is equal to 1, obtain a suitable standard vector based on the overall question vector.

[0122] When the number of subcategories is 1, it means that the missing word group category is the remaining subcategory content. Since the number of subcategory content of the missing word group category is 1, the moduli are the same, so the appropriate standard vector can be directly obtained based on the overall question vector.

[0123] Step 204: When the number of subcategories is greater than 1, the corresponding inquiry words are searched from a preset inquiry database based on the missing word group category.

[0124] The inquiry language is a language for asking which subcategory of the missing word group category the content belongs to. For example: the missing word group category is the object, such as: product, then the inquiry language is to ask which aspect of the product is related to the content. The database stores the mapping relationship between the missing word group category and the inquiry language. It is obtained after the staff in this field judges and identifies the subcategories that can be derived from the missing word group category, and then describes and records it according to their own experience and in accordance with their own identity experience. When the system receives the corresponding missing word group category, it automatically finds the corresponding inquiry language from the database and outputs it.

[0125] Step 205: Generate and output an adapted inquiry script based on the question content, subcategory content and inquiry script.

[0126] Adaptive inquiry words are generated based on the content of the question, which are consistent with the characteristics of the question content and easy to understand. The generation method is to first mention the question content and then form an adaptive inquiry wording.

[0127] Reference Figure 3 When the number of subcategories is greater than 1, the method for generating and outputting an adaptive inquiry speech based on the question content and the inquiry speech also includes:

[0128] Step 300: When the number of subcategories is greater than 1, obtain historical question content.

[0129] The historical question content is the content of the question that has been asked before. The method of obtaining it here can be to retrieve it from the storage space, that is, there is a database in the system that stores the historical question content in the past.

[0130] It should be noted that the historical problems formed in this place are short-term

[0131] Step 301: when there is historical question content, determine a keyword grouping category based on the historical question content, and define the keyword grouping category as a historical keyword grouping category.

[0132] The determination method is similar to step 102 and will not be described in detail here.

[0133] Step 302: Matching is performed based on the historical keyword group category and the missing keyword group category.

[0134] The matching method is text matching.

[0135] Step 303: When there is a historical keyword grouping category that matches the missing grouping category, the question keyword corresponding to the historical keyword grouping category is output as the question keyword corresponding to the missing grouping category, and the inquiry words are not searched.

[0136] When there is a historical keyword word group category that matches the missing word group category, it means that the content corresponding to the missing word group category is most likely unchanged, because most people will not repeat the corresponding content by default when the objects are the same, and when the objects are different, most people will say or write. Therefore, this time meets the previous situation of the same object, so the question keyword corresponding to the historical keyword word group category is output as the question keyword corresponding to the missing word group category, and the inquiry language is not searched.

[0137] Step 304: When there is no historical keyword word group category matching the missing word group category, an adaptive inquiry language is generated based on the question content, subcategory content and inquiry language and outputted.

[0138] When there is no historical keyword grouping category that matches the missing grouping category, it means that it has not been mentioned before and there is no reference basis, so an adaptive inquiry dialogue is generated based on the question content, subcategory content and inquiry dialogue and output.

[0139] Reference Figure 4, when there is no historical keyword word category matching the missing word category, the method for generating and outputting an adaptive inquiry speech based on the question content, the subcategory content and the inquiry speech includes:

[0140] Step 400: based on the keywords and keyword group categories corresponding to the historical keyword group categories, the subcategory contents are screened in sequence to obtain the number of screened subcategories, and the number of subcategories is defined as the number of screened subcategories.

[0141] The screening method here is to search from the database. That is, the database stores the mapping relationship between historical keyword grouping categories, keywords, keyword grouping categories and sub-category contents. The staff in this field determines the sub-category contents of the keyword grouping categories associated with each historical keyword grouping category and the corresponding keywords based on their own experience and common knowledge. When the system receives the corresponding historical keyword grouping categories, keywords, and keyword grouping categories, it automatically outputs the corresponding sub-category contents from the database, and counts the corresponding sub-category quantity, which is defined as the output of the filtered sub-category quantity.

[0142] Step 401: When the number of filtered subcategories is still greater than 1, continue to filter the subcategory content in sequence based on the keywords corresponding to the historical keyword grouping categories and the keywords corresponding to the keyword grouping categories.

[0143] When the number of filtered subcategories is still greater than 1, it means that the number of subcategories cannot be reduced to 1 through the historical keyword group category at this time, so continue to find a new historical keyword group category for filtering.

[0144] Step 402: When the number of subcategories is still greater than 1 after all the keywords corresponding to the historical keyword grouping categories and the keywords corresponding to the keyword grouping categories have been screened, the subcategory content corresponding to the current number of subcategories is defined as the screened subcategory content.

[0145] When all the keywords corresponding to the historical keyword group categories and the keywords corresponding to the keyword group categories have been screened and the number of subcategories is still greater than 1, it means that the number of subcategories cannot be reduced to 1 in the end, but the number of subcategories has been reduced. Therefore, at this time, you can generate sales talk through the reduced content.

[0146] Step 403: Generate and output an adapted inquiry script based on the question content, the filtered subcategory content and the inquiry script.

[0147] The method here is similar to step 205 and will not be described in detail here. The difference is that the method used here is to filter subcategory content.

[0148] Step 404: after filtering keywords corresponding to a number of historical keyword word group categories or keywords corresponding to keyword word group categories, when the number of filtered subcategories is equal to 1, define the subcategory content at this time as the only subcategory content.

[0149] When the number of filtered subcategories is equal to 1, it means that the number of subcategory contents at this time is 1, and it can be directly output as a missing word group category.

[0150] Step 405: Output the unique sub-category content as the question keyword corresponding to the missing word group category, and do not search for the inquiry words.

[0151] Reference Figure 5 When the number of subcategories is greater than 1, the method for generating and outputting an adaptive inquiry speech based on the question content and the inquiry speech also includes:

[0152] Step 500: When the missing word group category and the preset picture analyzable category are the same, an input question picture is obtained.

[0153] The analyzable image category is the category that can be analyzed through the image. It is set manually. The question image is the image that is manually input when the question content is raised. When the missing word category and the analyzable image category are the same, it means that it is possible to analyze through the image at this time, so it is necessary to know whether there is a question photo at this time.

[0154] Step 501: Matching is performed based on the problem image, the missing word group category and the preset analyzable features to obtain the actual analyzable features.

[0155] Analyzable features are features that can be analyzed in the image. The actual analyzable features are the features of the missing word group category in the problem image that meet the analyzable features. Here, the features analyzed in the problem image are first determined based on the analyzable features. The analysis method can be image comparison or analysis through feature analysis software. Then, it is sufficient to identify whether the subcategory content corresponding to the feature and the missing word group category represents the same. It should be noted that each actual analyzable feature also carries some information, including the corresponding feature name and status.

[0156] Step 502: Based on the actual analyzable features, corresponding analyzable feature keywords are searched from a preset keyword database.

[0157] The analyzable feature keyword is a keyword corresponding to the actual analyzable feature. The database stores the mapping relationship between the actual analyzable feature and the analyzable feature keyword, which is set by the staff in this field according to the image corresponding to the analyzable feature and the staff's experience. When the system receives the corresponding actual analyzable feature, it automatically finds the corresponding analyzable feature keyword from the database and outputs it.

[0158] Step 503: Outputting the analyzable characteristic keywords as question keywords corresponding to the missing word group category without searching for inquiry words.

[0159] Step 504: When the missing word group category and the image analyzable category are different, an adaptive inquiry language is generated based on the question content, subcategory content and inquiry language and outputted.

[0160] If they are not the same, it means that it cannot be confirmed through the picture, and you can only ask the person who asked the question.

[0161] Reference Figure 6 , and also includes a further after-sales customer service method when obtaining the input problem picture, the method includes:

[0162] Step 600: Matching is performed based on the problem image and preset analyzable features to obtain all analyzable features in the image.

[0163] The analyzable features in the graph are features that can be analyzed in the graph. This is the first step in step 501, and will not be described in detail here.

[0164] Step 601: Based on the analyzable features in the graph, corresponding analyzable feature keywords are searched from a keyword database.

[0165] The analyzable feature keywords in the graph are keywords of the features that can be analyzed in the graph. The search method is similar to step 502 and will not be described in detail here.

[0166] Step 602: Identify the question keywords based on the action words and state words in the specific identification word library to determine the in-graph keyword group category corresponding to the in-graph analyzable feature keywords.

[0167] The keyword group category in the graph is the group category corresponding to the analyzable characteristic keyword in the graph. The identification method is similar to step 102 and will not be repeated here.

[0168] Step 603: Based on the analyzable characteristic keywords in the graph, the corresponding in-graph modulus is searched from the database corresponding to the modulus database number.

[0169] The modulus in the graph is the modulus corresponding to the analyzable characteristic keyword in the graph. The search method is similar to steps 103-104 and will not be described in detail here.

[0170] Step 604: Determine the modulus within the graph and the modulus relationship between the moduli based on the keyword group word category within the graph and the keyword group word category.

[0171] The modulus relationship is the relationship between the modulus in the figure and the modulus of the same word group category. The method of determination is to subtract the modulus of the two and then compare them with a reference value. This reference value is set manually. If it is greater than this reference value, it means that the two are completely different. If it is less than this reference value, it means that the two are similar. If it is equal to 0, it means that the two are the same.

[0172] Step 605: No operation is performed when the modulus relationship is a preset similar relationship or the same relationship.

[0173] The similarity relationship is the relationship that the in-graph analyzable feature keywords corresponding to the in-graph keyword word group category and the question keywords corresponding to the keyword word group category have similar meanings. The same relationship is the relationship that the in-graph analyzable feature keywords corresponding to the in-graph keyword word group category and the question keywords corresponding to the keyword word group category have the same meanings.

[0174] If the modulus relationship is a preset similarity relationship or an identical relationship, it means that the content of the customer's problem description is similar or identical to the content shown in the picture, and there is no need to doubt it.

[0175] Step 606: When the modulus relationship is a preset complete distinction relationship, the corresponding in-graph keyword group category is defined as an abnormal keyword group category, and the in-graph analyzable feature corresponding to the abnormal keyword group category is defined as an abnormal feature.

[0176] The completely distinct relationship is a completely distinct relationship between the meanings of the analyzable feature keywords in the graph corresponding to the keyword word group category in the graph and the question keywords corresponding to the keyword word group category.

[0177] Step 607: Mark the area in the problem image corresponding to the abnormal feature, and output it with a preset warning signal.

[0178] The warning signal is to alert the user to the problem that there is a contradiction between this place and the keywords in the description. The purpose of marking the area in the problem picture corresponding to the abnormal feature is to remind the customer where the problem is.

[0179] Reference Figure 7 , and also includes an after-sales customer service method when the similarity corresponding to the question keyword or the suitable standard vector cannot be obtained and is less than a preset unconfirmable critical value, the method comprising:

[0180] Step 700: Decompose the question content based on reference words in a preset industry knowledge base to obtain some question keywords.

[0181] If the critical value cannot be confirmed, the corresponding similarity is too low and there is no match at all, that is, the question content is basically different from the questions stored in the database. Some question keywords are question keywords that can be disassembled in the question content. The disassembly method here is similar to step 101 and will not be repeated here.

[0182] Step 701: Based on some of the question keywords, the corresponding tag human service personnel are searched from the preset tag database.

[0183] Some of the question keywords are service personnel who are good at the field related to the some of the question keywords. The database stores the mapping relationship between some of the question keywords and the labeled manual service personnel. The staff in this field observes whether it is associated with some keywords according to the field corresponding to the actual service personnel. When there is no labeled manual service personnel, it is randomly sent to any service personnel. If the same part of the question keyword corresponds to multiple labeled manual service personnel, it is randomly sent to any labeled manual service personnel.

[0184] Step 702: Send the question content to the label manual service personnel and receive the manual reply content.

[0185] The manual reply content refers to the reply content provided by the label manual service personnel to the question content.

[0186] Step 703: Output the manual reply content, and obtain the corresponding question keywords, keyword group categories, modulus database numbers and moduli based on the manual reply content and the preset decomposition inquiry operation, and record them in the corresponding specific identification vocabulary and modulus database number.

[0187] The disassembly inquiry operation is an operation to inquire according to the question keywords, keyword word group categories, modulus database number and modulus order. The inquiry object here is the label manual service personnel, and the inquiry content is: what are the question keywords of this question content, what is the keyword word group category, what is the modulus database number, and what is the modulus.

[0188] Step 704: Store the manual reply content in the answer database.

[0189] The manual reply content, question keywords, keyword group category, modulus database number and modulus are stored here and associated.

[0190] Based on the same inventive concept, an embodiment of the present invention provides an intelligent after-sales customer service system.

[0191] Reference Figure 8 , an intelligent after-sales customer service system, comprising:

[0192] The acquisition module is used to obtain question content, question images and manual response content;

[0193] A memory, used for storing a program of a control method of an intelligent after-sales customer service method;

[0194] The program in the processor memory can be loaded and executed by the processor to implement a control method of an intelligent after-sales customer service method.

[0195] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device and unit described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0196] An embodiment of the present invention provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed by a method for intelligent after-sales customer service.

[0197] Computer storage media include, for example, various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.

[0198] Based on the same inventive concept, an embodiment of the present invention provides an intelligent terminal, including a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute an intelligent after-sales customer service method.

[0199] The above are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Any feature disclosed in this specification (including the abstract and drawings), unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

Claims

1. An intelligent after-sales customer service method, characterized in that: include: Get the input question content; Decompose the question content based on the reference words in the preset industry knowledge base to obtain the question keywords; Based on the action words and state words in the preset specific identification word library, the question keywords are identified to determine the keyword word group category corresponding to the question keywords, wherein the keyword word group category is the grammatical structure position or sentence component category of the vocabulary corresponding to the question keywords when forming a sentence; Determine the corresponding module database number based on the keyword group category; Based on the question keyword, the corresponding modulus is found from the database corresponding to the modulus database number, where the modulus is a numerical value in the direction corresponding to the question keyword, and the purpose is to express the corresponding keyword so that it becomes a vectorized vocabulary; Based on the modulus of the question keywords, an overall question vector is formed. The overall question vector is a vector corresponding to the entire question content. The questions in the imported content data are segmented, and the questions in the content data are vectorized according to the results of the segmentation. The segmentation results and importance are also included in the vectorized representation according to RF-IDF. Based on the overall problem vector and all preset standard vectors, a two-norm similarity calculation is performed to obtain a standard vector with the greatest similarity, and the standard vector is defined as a suitable standard vector; Find the corresponding appropriate reply content from the preset answer database based on the appropriate standard vector; Output the appropriate response content.

2. The intelligent after-sales customer service method according to claim 1, characterized in that: Also included is a method for obtaining a suitable standard vector based on the overall problem vector, the method comprising: Determine the missing word category based on matching the keyword word category and the preset required category; When the missing word category does not exist, a suitable standard vector is obtained based on the overall question vector; When the missing word group category exists, searching the corresponding subcategory quantity and subcategory content from a preset subcategory database based on the missing word group category; When the number of subcategories is equal to 1, a suitable standard vector is obtained based on the overall problem vector; When the number of subcategories is greater than 1, the corresponding inquiry words are searched from the preset inquiry database based on the missing word group category; Generate and output adapted inquiry dialogue based on question content, subcategory content and inquiry dialogue.

3. The intelligent after-sales customer service method according to claim 2, characterized in that: When the number of subcategories is greater than 1, the method for generating and outputting an adaptive inquiry speech based on the question content and the inquiry speech also includes: Get historical question content when the number of subcategories is greater than 1; When there is historical problem content, determining a keyword grouping category based on the historical problem content, and defining the keyword grouping category as a historical keyword grouping category; Matching based on historical keyword grouping categories and missing keyword grouping categories; When there is a historical keyword grouping category that matches the missing grouping category, the question keyword corresponding to the historical keyword grouping category is output as the question keyword corresponding to the missing grouping category, and the inquiry words are not searched; When there is no historical keyword group category that matches the missing group category, an adaptive inquiry dialogue is generated based on the question content, subcategory content and inquiry dialogue and output.

4. The intelligent after-sales customer service method according to claim 3, characterized in that: When there is no historical keyword word category matching the missing word category, the method for generating and outputting an adaptive inquiry speech based on the question content, the subcategory content and the inquiry speech includes: Based on the keywords and keyword group categories corresponding to the historical keyword group categories, the subcategory contents are screened in turn to obtain the number of screened subcategories, and the number of subcategories is defined as the number of screened subcategories; When the number of subcategories to be screened is still greater than 1, the subcategory contents are screened in sequence based on the keywords corresponding to the historical keyword grouping categories and the keywords corresponding to the keyword grouping categories; When the number of subcategories is still greater than 1 after all the keywords corresponding to the historical keyword grouping categories and the keywords corresponding to the keyword grouping categories have been screened, the subcategory content corresponding to the current number of subcategories is defined as the screened subcategory content; Generate and output the adapted inquiry script based on the question content, filtered subcategory content and inquiry script; After the keywords corresponding to several historical keyword word grouping categories or the keywords corresponding to the keyword word grouping categories are screened, when the number of screened subcategories is equal to 1, the subcategory content at this time is defined as the only subcategory content; The unique subcategory content is output as the question keyword corresponding to the missing word group category, and the inquiry language is not searched.

5. The intelligent after-sales customer service method according to claim 2, characterized in that: When the number of subcategories is greater than 1, the method for generating and outputting an adaptive inquiry speech based on the question content and the inquiry speech also includes: When the missing word group category and the preset picture analyzable category are the same, the input question picture is obtained; Matching is performed based on the problem image, the missing word group category and the preset analyzable features to obtain the actual analyzable features; Based on the actual analyzable features, corresponding analyzable feature keywords are searched from a preset keyword database respectively; Output the analyzable feature keywords as the question keywords corresponding to the missing word group category, without searching for the inquiry words; When the missing word group category and the image analyzable category are different, an adaptive inquiry dialogue is generated based on the question content, subcategory content and inquiry dialogue and output.

6. The intelligent after-sales customer service method according to claim 5, characterized in that: It also includes a further after-sales customer service method when obtaining the input problem picture, which includes: Matching the problem image with the preset analyzable features to obtain all the analyzable features in the image; Based on the analyzable features in the graph, corresponding analyzable feature keywords in the graph are searched from the keyword database respectively; Based on the action words and state words in the specific identification word library, the question keywords are identified to determine the keyword group category in the graph corresponding to the analyzable characteristic keywords in the graph; Based on the analyzable characteristic keywords in the graph, the corresponding modulus in the graph is searched from the database corresponding to the modulus database number; Determining a modulus relationship between the modulus within the graph and the modulus between the moduli based on the keyword group word category within the graph and the keyword group word category; When the modulus relationship is a preset similarity relationship or the same relationship, no operation is performed; When the modulus relationship is a preset complete distinction relationship, the corresponding in-graph keyword group category is defined as an abnormal keyword group category, and the in-graph analyzable feature corresponding to the abnormal keyword group category is defined as an abnormal feature; The area in the problem image corresponding to the abnormal features is marked and output with the preset warning signal.

7. The intelligent after-sales customer service method according to claim 1, characterized in that: The method also includes an after-sales customer service method when the similarity corresponding to the question keyword cannot be obtained or the suitable standard vector is less than a preset unconfirmable critical value, the method comprising: Disassemble the question content based on the reference words in the preset industry knowledge base to obtain some question keywords; Based on some of the question keywords, the corresponding label human service personnel are found from the preset label database; Send questions to label service personnel and receive manual replies; Output the manual reply content, and obtain the corresponding question keywords, keyword group categories, modulus database numbers and moduli based on the manual reply content and the preset disassembly inquiry operation, and record them in the corresponding specific identification word library and modulus database number; The manual responses are stored in the answer database.

8. An intelligent after-sales customer service system, characterized in that: include: The acquisition module is used to obtain question content, question images and manual response content; A memory, used to store a program of a control method of an intelligent after-sales customer service method according to any one of claims 1 to 7; The program in the processor memory can be loaded and executed by the processor to implement a control method of an intelligent after-sales customer service method as described in any one of claims 1 to 7.

9. Intelligent terminal, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute an intelligent after-sales customer service method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that A computer program is stored which can be loaded by a processor and execute an intelligent after-sales customer service method as claimed in any one of claims 1 to 7.

Images