Business problem scenario identification method and device, electronic equipment and storage medium

Abstract

The application provides a service problem scenario recognition method and device, electronic equipment and a storage medium, comprising: determining a problem complaint text, preprocessing the problem complaint text to obtain a to-be-matched word sequence; and calculating the matching degree of the to-be-matched word sequence based on a scenario word matrix to obtain a scenario recognition output result. The application obtains more feature descriptions under mobile service problem scenarios by using word distribution relationship representation learning under a neural network training model, and realizes greater range coverage of scenario and text description in combination with a word description relationship matrix, thereby improving the scenario recognition capability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent operation and maintenance, and particularly relates to a business problem scenario identification method and device, electronic equipment and a storage medium. BACKGROUND

[0002] In a daily business scenario, when a business operator encounters a problem and needs to feedback, the business operator usually interacts with a customer service robot and feeds back the problem, the customer service robot analyzes and processes the feedback problem, and then outputs a corresponding answer.

[0003] Generally, the customer service robot performs text segmentation on the input problem content through a jieba tool, converts the word sequence into a word vector by using a digitized array representation, and then performs similarity calculation on the text words by using a cosine function, finds related text knowledge in a historical knowledge base according to the calculation result, and realizes associated search on the text content.

[0004] In the above scheme, the positioning of the problem and the provision of the solution are both knowledge searches conforming to the word description of the word segmentation result of a text content, and the scene screening and identification has certain limitations, especially for different text descriptions in the same scene, which cannot be identified and a solution cannot be provided, and the scene coverage is limited. SUMMARY

[0005] The present application provides a business problem scenario identification method and device, electronic equipment and a storage medium, to solve the defect that the existing technology has limitations in outputting a corresponding solution to a problem.

[0006] In a first aspect, the present application provides a business problem scenario identification method, comprising:

[0007] determining a problem complaint text, preprocessing the problem complaint text to obtain a to-be-matched word sequence;

[0008] calculating the matching degree of the to-be-matched word sequence based on a scene word matrix to obtain a scene identification output result.

[0009] In one embodiment, the determination of the problem complaint text, the preprocessing of the problem complaint text to obtain the to-be-matched word sequence specifically comprises:

[0010] obtaining problem description text content;

[0011] extracting key words in the problem description text content based on a gensim framework and jieba segmentation to obtain the to-be-matched word sequence.

[0012] In one embodiment, the scene word matrix is obtained by the following steps:

[0013] obtain historical complaint work order data, perform text extraction on the historical complaint work order data, store preset field set information, and construct a vocabulary dictionary;

[0014] The relationship between the vocabulary sequence and the vocabulary is trained and learned using a preset word vector model to obtain the scene vocabulary matrix.

[0015] In one embodiment, the obtaining historical complaint work order data, performing text extraction on the historical complaint work order data, storing preset field set information, and constructing a vocabulary dictionary specifically include:

[0016] The business category, problem attribute, and work order keyword in the historical complaint work order data are stored in a structured database.

[0017] The business category, problem attribute, and work order keyword are mapped to vocabulary attributes to obtain the preset field set information.

[0018] The preset field set information is stored in a text file of the vocabulary dictionary and is entered and marked row by row according to the vocabulary attribute mapping label.

[0019] The vocabulary description record of the tool is determined and is marked using the vocabulary attribute mapping label.

[0020] In one embodiment, the training and learning of the relationship between the vocabulary sequence and the vocabulary using a preset word vector model to obtain the scene vocabulary matrix specifically include:

[0021] The input of the CBOW model includes business category description vocabulary, problem attribute description vocabulary, and first keyword description vocabulary, and the output includes tool description words corresponding to the result with the maximum possibility. The word vector set corresponding to the business category description vocabulary, the problem attribute description vocabulary, the first keyword description vocabulary, and the tool description words is obtained using a DNN back propagation algorithm. The input of the CBOW model also includes second keyword description vocabulary.

[0022] The historical problem work order in a preset problem scene is manually marked, and the tool used in the preset problem scene is field supplemented and marked to obtain model training data. A logarithmic likelihood function is used as an objective function, and the scene vocabulary matrix is trained based on the word vector set.

[0023] In one embodiment, the matching degree calculation of the to-be-matched vocabulary sequence based on the scene vocabulary matrix to obtain the scene recognition output result specifically includes:

[0024] According to the order of business category attribution, problem attribute weight and keyword description, the matching degree of the problem scene is calculated in sequence to obtain an initial recognition result;

[0025] The initial recognition result is subjected to slot matching degree calculation to obtain the scene recognition output result.

[0026] In one embodiment, the scene vocabulary matrix is used to calculate the matching degree of the to-be-matched vocabulary sequence to obtain the scene recognition output result, and then the following steps are further included:

[0027] According to the scene recognition output result, a corresponding solution tool ID is obtained, a recommended tool service is obtained from the solution tool ID, and a problem diagnosis result is obtained for the user.

[0028] In a second aspect, the application further provides a business problem scene recognition device, which comprises:

[0029] A preprocessing module is configured to determine a problem complaint text, pre-process the problem complaint text, and obtain a to-be-matched vocabulary sequence.

[0030] A calculation module is configured to calculate the matching degree of the to-be-matched vocabulary sequence based on a scene vocabulary matrix to obtain a scene recognition output result.

[0031] In a third aspect, the application further provides an electronic device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the business problem scene recognition method according to any one of the above aspects when executing the program.

[0032] In a fourth aspect, the application further provides a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the steps of the business problem scene recognition method according to any one of the above aspects.

[0033] The business problem scene recognition method, device, electronic device and storage medium provided by the application can obtain more feature descriptions in the mobile business problem scene by using the vocabulary distribution relationship representation learning under the neural network training model, and can realize a larger range of coverage of the scene and the text description in combination with the vocabulary description relationship matrix, thereby improving the scene recognition capability. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and those skilled in the art can obtain other drawings according to these drawings without any creative effort.

[0035] Figure 1 is a flowchart of a business problem scenario recognition method provided by the present application;

[0036] Figure 2 is a schematic diagram of a business problem scenario diagnosis principle provided by the present application;

[0037] Figure 3 is a CBOW model schematic diagram provided by the present application;

[0038] Figure 4 is a structural schematic diagram of a business problem scenario recognition device provided by the present application;

[0039] Figure 5 is a structural schematic diagram of an electronic device provided by the present application. DETAILED DESCRIPTION

[0040] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0041] In view of the limitations of problem positioning and corresponding solution providing in the prior art, the present application provides a business problem scenario recognition method, Figure 1 is a flowchart of a business problem scenario recognition method provided by the present application, as Figure 1 shown, comprising:

[0042] 101, determining a problem complaint text, pre-processing the problem complaint text to obtain a to-be-matched word sequence;

[0043] 102, calculating the matching degree of the to-be-matched word sequence based on a scenario word matrix, to obtain a scenario recognition output result.

[0044] Specifically, for the to-be-recognized problem complaint text, the present application obtains the to-be-matched word sequence after pre-processing the problem complaint text, and then calculates the matching degree of the to-be-matched word sequence by the constructed scenario word matrix, to obtain the scenario recognition output result.

[0045] It can be understood that the present application is based on mobile company business rule data, combined with the historical problem solving experience of business experts, the business rules and the relationship logic between businesses in the actual business handling process are clarified, and the intelligent diagnosis platform is integrated, the intelligent diagnosis of business problem guarantee is realized, the effectiveness of problem processing is improved, and the business pressure of background operation and maintenance personnel is relieved. Support the query diagnosis of user's main personal basic business, personal bill, various types of package usage and billing priority and other business problems. Through the input of the problem description of the front-end operator in the intelligent assistant interaction interface, the system carries out intelligent scene recognition, recommends self-service tools, the operator clicks the tool to enter the diagnosis interface, the system automatically carries the customer problem number, clicks the query to obtain various types of resource states of the current business of the user, and the problem reason analysis and solution operation prompt.

[0046] The present application realizes the classification of scenes and diverse text descriptions by training different text descriptions under the same scene based on a neural network, realizes the representation of different vocabulary descriptions under a type of scene through a scene description vocabulary matrix, and realizes scene recognition and prediction of problem description text based on a scene description matrix, thereby improving the scene recognition capability.

[0047] Based on the above embodiment, step 101 specifically includes:

[0048] The business category, problem attribute and work order keyword in the historical complaint work order data are stored in a structured database;

[0049] The vocabulary attribute mapping is performed on the business category, the problem attribute and the work order keyword to obtain the preset field set information;

[0050] The preset field set information is stored in a text file of the vocabulary dictionary, and is entered and marked row by row according to the vocabulary attribute mapping label;

[0051] The vocabulary description record of the tool is determined, and the vocabulary attribute mapping label is used for marking.

[0052] Specifically, first, the historical business name is screened and extracted, a vocabulary dictionary chinamobilewords.txt is created, the historical complaint work order data is textually extracted, the business category, problem attribute and work order keyword field are stored, the above contents are stored in a structured database; then the business category, problem attribute, keyword description vocabulary attribute are mapped, the contents of each field are converted to the chinamobilewords.txt file, the label mapping is performed according to the vocabulary attribute and is entered and marked row by row, and the vocabulary of the tool is described and recorded, and the toolkey is marked correspondingly.

[0053] Here, the stored content can refer to the following commonly used format:

[0054] Opening an account business

[0055] Closing an account business

[0056] Transferring an account business

[0057] Opening an account verification property

[0058] Identity verification property

[0059] Exceeding the limit servicekey

[0060] One certificate multiple number servicekey

[0061] Five full number servicekey

[0062] Full network one certificate multiple number query toolkey

[0063] Opening one certificate multiple number verification toolkey

[0064] Recharge card query toolkey

[0065] …

[0066] The application divides the mobile service field vocabulary into four vocabulary attributes of business, property, servicekey and toolkey, trains the relationship between business, property, servicekey and toolkey vocabulary in the typical problem scene based on historical complaint work order data using a neural network model, obtains different vocabulary descriptions in the same diagnostic scene, and represents the scene vocabulary relationship through a scene feature word matrix, thereby realizing the relationship mapping of the diagnostic scene and the related problem description.

[0067] Based on any of the above embodiments, the scene vocabulary matrix is obtained by the following steps:

[0068] Obtain historical complaint work order data, perform text extraction on the historical complaint work order data, store preset field set information, and construct a vocabulary dictionary;

[0069] Train and learn the relationship representation between the vocabulary sequence and the vocabulary using a preset word vector model to obtain the scene vocabulary matrix.

[0070] The obtaining historical complaint work order data, text extraction on the historical complaint work order data, storage of preset field set information, construction of a vocabulary dictionary, specifically includes:

[0071] The business category, problem attribute and work order keyword in the historical complaint work order data are stored into a structured database.

[0072] The vocabulary attribute mapping is performed on the business category, the problem attribute and the work order keyword, and the preset field set information is obtained.

[0073] The preset field set information is stored into a text file of the vocabulary dictionary, and is entered and marked row by row according to the vocabulary attribute mapping label.

[0074] The vocabulary description record of the tool is determined, and the vocabulary attribute mapping label is marked.

[0075] The adopting a preset word vector model to train and learn the relationship representation between the word sequence and the word, and obtaining the scene vocabulary matrix, specifically includes:

[0076] The input of the CBOW model includes business class description words, problem attribute description words and first keyword description words, and the output includes tool description words corresponding to the result with the maximum possibility, and the DNN back propagation algorithm is adopted to obtain the word vector set corresponding to the business class description words, the problem attribute description words, the first keyword description words and the tool description words; wherein the input of the CBOW model also includes second keyword description words.

[0077] The historical problem work order in a preset problem scene is manually marked, and the tool used in the preset problem scene is field supplemented and marked, to obtain model training data, the log likelihood function is used as the objective function, and the scene vocabulary matrix is trained based on the word vector set.

[0078] Specifically, the scene vocabulary matrix proposed in the application is obtained by the following steps: Figure 2

[0079] The CBOW (Continuous Bag-of-Words) model is used to learn the relationship representation between the word sequence and the word, and the training input of the CBOW model in the application is the word vector corresponding to the context-related word pair in the feature scene, and the corresponding output is the word vector of the scene description word, that is, the tool associated with the scene description word, and the schematic diagram of the CBOW model is shown in Figure 3 .

[0080] ​The context size of the model input is selected as 4, which includes the corresponding 3-dimensional description vocabulary, i.e. business category, problem attribute, keyword description 1, keyword description 2 (this slot can be empty, i.e. optional), and the output is the tool description word corresponding to the maximum calculation result possibility. The CBOW neural network model input layer has 4 neurons and the output layer has 1 neuron. Through the back propagation algorithm of DNN, the parameters of the DNN model can be solved, and the word vectors corresponding to all words are obtained at the same time. The training goal is to maximize the softmax probability corresponding to the specific word to be trained, that is, the relevant scene vocabulary.

[0081] Here, 10 problem scenarios are taken as examples, each of which can be processed by a diagnostic tool, such as "full network one certificate multi-number query", "recharge card query", "short message sending state query", "out-of-plan traffic query", etc. Through manual marking of historical problem work order problem scenarios, work order data corresponding to problem scenarios that can be solved by diagnostic tools are found, and fields of tools that can be used for work order corresponding scenarios are supplemented and marked. Then we mark and extract work orders belonging to the above 10 types of problem scenarios in historical complaint work orders, and supplement the field of tool names mapped by each work order scenario at the same time. The extracted and marked work order set is used as training data for subsequent learning and use. In addition, the objective function of learning adopts the log-likelihood function:

[0082]

[0083] Suppose the word vector space dim is V, the context word number is C, and w represents any word in the corpus C:

[0084]

[0085] i w represents the index of word w in the dictionary, y w represents the corresponding probability set of each feature scene word.

[0086] Through the historical data of historical work orders and their corresponding problem scenario positioning, the representation relationship learning of word description and scene description in the mobile business field is carried out, the mapping of key words and problem scenarios is realized, and the word description matrix of each scene in 3 dimensions is obtained. Taking the "full network one certificate multi-number query" scenario as an example, its tool ID is 001, and its description matrix is as follows:

[0087] {001 open account identity verification one certificate multi-number data query

[0088] 001 open account identity verification over-limit multi-number

[0089] 001 open account customer data 5 numbers over-limit

[0090] 002Opening package verification package handling do as I please package

[0091] The first column is the tool number, the second column is the business category slot information, the third column is the problem attribute slot information, the fourth column is the keyword 1 slot information, and the fifth column is the keyword 2 slot information.

[0092] According to the scene description matrix, the corresponding "full-network multi-number query" diagnostic tool can be found by combining and mapping "opening verification", "one certificate multi-number", "over-limit", "data query" and the like, so that when there is a new task demand, the key word content in the description is extracted, the corresponding most possible output scene is found according to the description matrix distribution, and the related tool is found to be recommended to the user.

[0093] The scene word matrix is obtained by model training, which provides accurate and efficient word reference for subsequent scene recognition.

[0094] Based on any of the above embodiments, the method specifically comprises the following steps.

[0095] The matching degree of the problem scene is calculated in the order of business category, problem attribute weight and keyword description, and an initial recognition result is obtained.

[0096] The initial recognition result is subjected to slot matching degree calculation, and the scene recognition output result is obtained.

[0097] Specifically, according to the problem scene of the explicit solution, the problem positioning function of the problem scene is realized through software development, the number of the user is queried, the number state, the package attribution, the personal account amount information, the attribution information and the like of the user are queried, the corresponding business rule verification is combined, the place of business conflict is found. After finding the reason of the business conflict, the processing opinion is fed back in combination with the processing scheme, and the processing opinion is displayed in the foreground interface, so that the user can read the content.

[0098] For example, in the user number opening verification scene, the user number is first obtained, the current account attribution identity card information of the user is queried according to the user number, the number under the card is found through the card number, there can be one or more, if there are multiple, it is judged whether there are five, if there are five, the user is prompted that the number under the card has five, and the number opening is not allowed; if the number under the card of the user does not exceed five, the account state under each number is further judged, whether each card has no overdue record, if there is an overdue record, the user is prompted to pay the fee of other numbers, and the card attribution state is updated.

[0099] The system extracts the user's problem description text content in the operation and maintenance assistant, uses the gensim framework and jieba word segmentation to extract the key words of the text content, forms a key word sequence, the word sequence includes business category description, work order keyword description and other content, and predicts the probability of the scene through the scene description matrix word distribution, and the maximum probability value is the system predicted problem scene. Find the corresponding tool solution ID and recommend related tool services to the user for diagnosis.

[0100] The user's problem description text in the operation and maintenance assistant is "User 12345678911 wants to open a new mobile phone card, and the identity verification prompt cannot be opened during the opening of the account. Customer data query is required." Extract "opening of account", "identity verification" and "data query" as key words. "Opening of account" is a business category, "identity verification" is a problem attribute, and "data query" is a keyword description. The element feature description of four slot information distribution is "opening of account", "identity verification", "data query" and "" (i.e. null).

[0101] Matching degree calculation

[0102] In the identification of problem scenarios, the business category has the highest influence degree. First, find its business category attribution, and then calculate the slot matching degree under the condition that the business category is satisfied. In the calculation process, consider the weight distribution of each slot. The problem attribute weight is given priority, and the keyword description is given second priority. The weight factor θ of the problem attribute is 1 / 2, 1 / 4 and 1 / 4. The slot matching value p is p=0 (under the condition of slot matching) and p=1 (under the condition of slot mismatch).

[0103] The slot matching degree S is calculated as

[0104]

[0105] According to the business category slot value "opening of account", find the matrix subset W in the second column of the description matrix, and then calculate the matching degree according to the subset W. The problem attribute slot value is "identity verification", the keyword 1 description slot value is "data query", and the keyword 2 description slot value is "" (i.e. null).

[0106] Then S1 max=1 / 2+1 / 4=3 / 4, and the corresponding tool code ID is found to be 001. Then find the corresponding "all-network one certificate multi-number query" tool for the user.

[0107] The application is aimed at the segmented sequence of text processing, extracts key words, and calculates the matching degree of the output text and the text content in the historical knowledge base based on the scene feature matrix and the slot matching degree algorithm, obtains the maximum matching calculation result, realizes scene recognition prediction, and outputs more accurate scene recognition results.

[0108] Based on any of the above embodiments, the method further comprises, after step 102:

[0109] According to the scene recognition output result, the corresponding solution tool ID is obtained, the recommended tool service is obtained from the solution tool ID, and the user obtains the problem diagnosis result.

[0110] Specifically, after obtaining the business scene recognition result, the application identifies the mobile user number information by means of automation, checks the association relationship based on business rules for the personal business package, account attribute and the like, provides reason description for reasons such as business conflict or inconsistent business attribute state, realizes problem diagnosis, changes the state by one key for the scene that can perform data synchronization update, and realizes one-key repair of the problem.

[0111] The application realizes one-key automatic diagnosis of the problem scene by the business scene problem diagnosis device, and improves the problem processing efficiency.

[0112] The business problem scene recognition device provided by the application is described below, and the business problem scene recognition device described below can be correspondingly referred to the business problem scene recognition method described above.

[0113] Figure 4 The application provides a structure diagram of the business problem scene recognition device, as shown in Figure 4 The application provides a structure diagram of the business problem scene recognition device, as shown in

[0114] The preprocessing module 41 is used for determining a problem complaint text, preprocessing the problem complaint text, and obtaining a to-be-matched word sequence; and the calculation module 42 is used for calculating the matching degree of the to-be-matched word sequence based on a scene word matrix, and obtaining a scene recognition output result.

[0115] The application realizes the classification of scenes and diverse text descriptions by training different text descriptions under the same scene based on a neural network, realizes the representation of different word descriptions under a class of scenes by representing the relationship between scenes and word description sets through a scene description word matrix, realizes scene recognition prediction of problem description texts based on a scene description matrix, and improves the scene recognition capability.

[0116] Based on any of the above embodiments, the preprocessing module 41 is specifically used for:

[0117] Obtain the problem description text content; extract key words from the problem description text content based on the gensim framework and jieba word segmentation to obtain the word sequence to be matched.

[0118] Based on any of the above embodiments, the scene vocabulary matrix in the calculation module 42 is obtained through the following steps:

[0119] Obtain historical complaint work order data, extract text from the historical complaint work order data, store preset field set information, and construct a vocabulary dictionary;

[0120] The scene vocabulary matrix is ​​obtained by training and learning the relationship representation between the word sequence and the words using a preset word vector model.

[0121] The step of obtaining historical complaint work order data, extracting text from the historical complaint work order data, storing preset field set information, and constructing a vocabulary dictionary specifically includes:

[0122] Store the business category, problem attribute, and work order keyword from the historical complaint work order data into a structured database;

[0123] Vocabulary attribute mapping is performed on the business category, the problem attribute, and the work order keyword to obtain the preset field set information;

[0124] The preset field set information is stored in the text file of the vocabulary dictionary, and then entered and marked line by line according to the vocabulary attribute mapping tags;

[0125] The vocabulary description records of the tool are determined and labeled using the vocabulary attribute mapping tags.

[0126] Specifically, the step of training and learning the relationship representation between word sequences and words using a preset word vector model to obtain the scene word matrix includes:

[0127] The input of the CBOW model includes business category description words, problem attribute description words, and first keyword description words. The output includes the tool description words corresponding to the most probable results. The DNN backpropagation algorithm is used to obtain the word vector set corresponding to the business category description words, the problem attribute description words, the first keyword description words, and the tool description words. The input of the CBOW model also includes second keyword description words.

[0128] Historical problem work orders in the preset problem scenarios are manually marked, and the tools used in the preset problem scenarios are marked with additional fields to obtain model training data. The log-likelihood function is used as the objective function, and the scenario vocabulary matrix is ​​trained based on the word vector set.

[0129] Based on any of the above embodiments, the calculation module 42 is specifically used for:

[0130] The matching degree of the problem scenario is calculated sequentially according to the business category, problem attribute weight, and keyword description to obtain the initial recognition result; the slot matching degree is calculated on the initial recognition result to obtain the scene recognition output result.

[0131] Based on any of the above embodiments, a diagnostic module 43 is further included, wherein the diagnostic module 43 is specifically used for:

[0132] Based on the scene recognition output, the corresponding solution tool ID is obtained, and the recommended tool service is obtained from the solution tool ID, so that the user can obtain the problem diagnosis result.

[0133] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540. The processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a business problem scenario recognition method. This method includes: determining a problem complaint text; preprocessing the problem complaint text to obtain a sequence of words to be matched; and calculating the matching degree of the sequence of words to be matched based on a scenario word matrix to obtain a scenario recognition output result.

[0134] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0135] In another aspect, the present application also provides a computer program product, which comprises a computer program stored on a non-transitory computer-readable storage medium, and the computer program comprises program instructions, when the program instructions are executed by a computer, the computer can execute the business problem scenario recognition method provided by the above method, and the method comprises the following steps: determining a problem complaint text, pre-processing the problem complaint text to obtain a to-be-matched word sequence; and performing matching degree calculation on the to-be-matched word sequence based on a scenario word matrix to obtain a scenario recognition output result.

[0136] In another aspect, the present application also provides a non-transitory computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the business problem scenario recognition method provided by the above method, and the method comprises the following steps: determining a problem complaint text, pre-processing the problem complaint text to obtain a to-be-matched word sequence; and performing matching degree calculation on the to-be-matched word sequence based on a scenario word matrix to obtain a scenario recognition output result.

[0137] The device embodiments described above are only schematic, wherein the units illustrated as separate components can or can not be physically separate, and the components illustrated as units can or can not be physical units, i.e., can be located in one place or distributed on a plurality of network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment. Those skilled in the art can understand and implement without creative labor.

[0138] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software plus necessary general hardware platforms, and of course can also be implemented by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0139] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for business problem scenario identification, characterized in that, The method comprises the following steps: determining a problem complaint text, preprocessing the problem complaint text, and obtaining a to-be-matched word sequence; calculating the matching degree of the to-be-matched word sequence based on a scene word matrix to obtain a scene recognition output result; The scene word matrix is obtained by the following steps: obtaining historical complaint work order data, extracting text from the historical complaint work order data, storing preset field set information, and constructing a word dictionary; training and learning the relationship representation between the word sequence and the word by using a preset word vector model to obtain the scene word matrix; The training and learning of the relationship representation between the word sequence and the word by using the preset word vector model to obtain the scene word matrix specifically comprises: determining that the input of the CBOW model includes business class description words, problem attribute description words and first keyword description words, and the output includes tool description words corresponding to the result with the maximum possibility, and obtaining a word vector set corresponding to the business class description words, the problem attribute description words, the first keyword description words and the tool description words by using a DNN back propagation algorithm; wherein the input of the CBOW model further includes second keyword description words; manually marking historical problem work orders in a preset problem scene, and supplementarily marking fields of tools used in the preset problem scene to obtain model training data, using a log likelihood function as a target function, and training the scene word matrix based on the word vector set; The first column of the scene word matrix is a tool number, the second column is a business category slot information, the third column is a problem attribute slot information, the fourth column is a first keyword slot information, and the fifth column is a second keyword slot information. The matching degree calculation of the to-be-matched word sequence based on the scene word matrix to obtain the scene recognition output result specifically comprises: sequentially calculating the matching degree of the problem scene according to the order of business category attribution, problem attribute weight and keyword description to obtain an initial recognition result; performing slot matching degree calculation on the initial recognition result to obtain the scene recognition output result; The slot matching degree calculation comprises: finding a matrix subset in which the second column is the business category slot value according to the business category slot value, and then performing problem attribute slot value and keyword slot value matching degree calculation according to the subset.

2. The business problem scenario recognition method of claim 1, wherein, The determination of the problem complaint text, the preprocessing of the problem complaint text, and the obtaining of the to-be-matched word sequence specifically comprise: obtaining problem description text content; extracting key words in the problem description text content based on a gensim framework and a jieba word segmentation to obtain the to-be-matched word sequence.

3. The business problem scenario recognition method of claim 1, wherein, The obtaining of the historical complaint work order data, the text extraction from the historical complaint work order data, the storage of the preset field set information, and the construction of the word dictionary specifically comprise: storing the business category, the problem attribute and the work order keyword in the historical complaint work order data into a structured database; performing word attribute mapping on the business category, the problem attribute and the work order keyword to obtain the preset field set information; The preset field set information is stored in a text file of the vocabulary dictionary, and is recorded and marked line by line according to the vocabulary attribute mapping label; The vocabulary description record of the tool is determined, and the vocabulary attribute mapping label is marked.

4. The business problem scenario recognition method of claim 1, wherein, The scene vocabulary matrix is used to calculate the matching degree of the to-be-matched vocabulary sequence, and a scene recognition output result is obtained. According to the scene recognition output result, a corresponding solution tool ID is obtained, and a recommended tool service is obtained from the solution tool ID, so that a user obtains a problem diagnosis result.

5. The service problem scenario recognition apparatus, characterized by, It comprises: A preprocessing module is configured to determine a problem complaint text, pre-process the problem complaint text, and obtain a to-be-matched vocabulary sequence; A calculation module is configured to calculate the matching degree of the to-be-matched vocabulary sequence based on a scene vocabulary matrix, and obtain a scene recognition output result. The calculation module is specifically configured to: Obtain historical complaint work order data, extract text from the historical complaint work order data, store preset field set information, and construct a vocabulary dictionary; A preset word vector model is used to train and learn the relationship representation between the vocabulary sequence and the vocabulary, and the scene vocabulary matrix is obtained. The preset word vector model is used to train and learn the relationship representation between the vocabulary sequence and the vocabulary, and the scene vocabulary matrix is obtained, specifically including: The input of the CBOW model includes business class description vocabulary, problem attribute description vocabulary and first keyword description vocabulary, and the output includes tool description words corresponding to the maximum likelihood result, and the word vector set corresponding to the business class description vocabulary, the problem attribute description vocabulary, the first keyword description vocabulary and the tool description word is obtained by using the DNN back propagation algorithm; wherein the input of the CBOW model further includes second keyword description vocabulary; The historical problem work order in the preset problem scene is manually marked, and the tool used in the preset problem scene is field supplemented and marked to obtain model training data, and the log likelihood function is used as the objective function, and the scene vocabulary matrix is trained based on the word vector set; The first column of the scene vocabulary matrix is the tool number, the second column is the business category slot information, the third column is the problem attribute slot information, the fourth column is the first keyword slot information, and the fifth column is the second keyword slot information. The scene vocabulary matrix is used to calculate the matching degree of the to-be-matched vocabulary sequence, and a scene recognition output result is obtained. The matching degree of the problem scene is calculated in the order of business category attribution, problem attribute weight and keyword description, and an initial recognition result is obtained. The initial recognition result is subjected to slot matching degree calculation to obtain the scene recognition output result. The slot matching degree calculation includes: According to the business category slot value, find the matrix subset in which the second column is the business category slot value, and then perform problem attribute slot value and keyword slot value matching degree calculation according to the subset.

6. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the steps of the business problem scene recognition method according to any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the steps of the business problem scenario identification method according to any one of claims 1 to 4.

Images