AI-based intelligent IT service request classification and distribution method and system

Abstract

The invention discloses an intelligent IT service request classification and distribution method and system based on AI. The method comprises the steps of performing semantic analysis on a service request, and extracting key information; classifying the service requests based on the semantic analysis result; evaluating the service emergency degree of the service request in combination with a semantic analysis result; matching a proper technical support object according to the content and the emergency degree of the service request; and finally, according to a matching result and an emergency degree evaluation result, distributing the service request to a corresponding technical support object for processing. According to the method, automatic and precise classification and distribution of the IT service requests can be realized, the service response efficiency and the processing effect are improved, and the method is suitable for a large-scale IT service scene.

Description

Technical Field

[0001] The present invention relates to the field of information technology service management technology, and in particular to an AI-based intelligent IT service request classification and allocation method and system. Background Art

[0002] As the complexity of enterprise IT systems continues to increase, the number and diversity of IT service requests are also growing exponentially, placing higher efficiency demands on IT support departments. The traditional IT service request processing process relies primarily on manual operations, including request reception, classification, priority determination, and personnel allocation. This is not only time-consuming and labor-intensive, but is also prone to human factors leading to problems such as misclassification and improper allocation, reducing response speed and service quality. Although some intelligent solutions currently on the market attempt to improve this process through technical means, they still have significant shortcomings in the depth of natural language understanding, multimodal input processing, contextual awareness, domain adaptability, and matching algorithms between technicians and requests. In particular, in terms of the degree of automation in knowledge base construction and updating, and the system's learning capabilities, existing technologies cannot yet meet the needs of efficient and accurate IT service management. Summary of the Invention

[0003] The purpose of this invention is to provide an AI-based intelligent IT service request classification and allocation method and allocation system to improve processing efficiency and service quality.

[0004] To solve the above technical problems, the present invention provides an AI-based intelligent IT service request classification and allocation method, which includes:

[0005] Perform semantic analysis on service requests;

[0006] classifying the service request based on the result of the semantic analysis;

[0007] Performing a service urgency assessment on the service request based on the semantic analysis result;

[0008] Matching the service request with a technical support object;

[0009] The service request is assigned to a corresponding technical support object according to the matching result and the evaluation result of the service urgency.

[0010] Furthermore, semantic analysis of service requests includes:

[0011] After converting the service request into a request text, performing text preprocessing on the request text;

[0012] Extract key feature vectors from the preprocessed request text or knowledge base.

[0013] Furthermore, extracting key feature vectors from the pre-processed request text or knowledge base includes at least one of the following steps:

[0014] Extracting words from the request text and obtaining a word vector set;

[0015] Extracting keywords from the request text and obtaining a keyword vector set;

[0016] Aggregating the word vector and the keyword vector into a grammatical structure vector according to the grammatical structure input by the user;

[0017] According to the semantics of the text input by the user, the word vector, the keyword vector and the grammatical structure vector are aggregated into a text vector.

[0018] Furthermore, extracting key feature vectors from the pre-processed request text or knowledge base further includes:

[0019] Find and retrieve historical questions related to the user's current question;

[0020] Searching and obtaining historical user information and historical device information associated with the historical information and user information mentioned in the request text;

[0021] Find and obtain the fault type and fault solution associated with the current problem of the user.

[0022] Furthermore, the text preprocessing includes at least one of text cleaning, text segmentation, part-of-speech tagging, and stop word filtering.

[0023] Furthermore, performing service urgency assessment based on the semantic analysis result includes:

[0024] According to a preset priority judgment rule, the service request is evaluated for service urgency based on the key feature vector;

[0025] The priority judgment rules are:

[0026] P = w1·R+w2·C+w3·I;

[0027] Among them, P is the comprehensive priority score, R is the real-time score, C is the processing complexity score, I is the impact score, and w1, w2, and w3 are the weight coefficients of each score.

[0028] Furthermore, classifying the service request based on the result of the semantic analysis includes:

[0029] Mapping the key feature vector to a category probability distribution of the service request;

[0030] Obtaining a category of the service request according to the category probability distribution;

[0031] The calculation method of the category probability distribution is:

[0032]

[0033] Where P(y=j|x) represents the probability that the input vector x belongs to category j, x is the input vector, w j is the weight vector of category j, K is the total number of categories, Represents the input vector x and the category j weight vector w j The dot product of .

[0034] Furthermore, according to the matching result, allocating the service request to a corresponding technical support object includes:

[0035] Calculating the similarity between the request text and the technical support object skill description text to obtain a content matching score;

[0036] The similarity is calculated as follows: similarity(A,B)=(A·B) / (||A||·|B||);

[0037] Where A and B are texts represented by vectors; ||A|| and ||B|| are the vector norms of each;

[0038] Calculate the correlation between service requests and technical support objects to obtain collaborative filtering scores;

[0039] The collaborative filtering score is calculated as follows:

[0040] Among them, similarity(i,j) represents the collaborative filtering score, R i and R j Represent the ratings or interaction behavior vectors of user i and user j in historical tickets respectively;

[0041] Performing weighted fusion on the content matching score and the collaborative filtering score to generate a comprehensive matching score;

[0042] The technical support object is matched with the service request according to the comprehensive matching score.

[0043] Furthermore, the method further comprises:

[0044] Calculate the TF-IDF value of each word in the request text;

[0045] extracting subject information from the request text;

[0046] Selecting key information in the request text based on the TF-IDF value and the relevance between the key information and the topic information;

[0047] and storing the key information;

[0048] The TF-IDF value is calculated as follows:

[0049] TF-IDF(t,d)=TF(t,d)·IDF(t);

[0050] Among them, TF(t,d) is the term frequency of word t in document d, and IDF(t) is the inverse document frequency of word t.

[0051] Furthermore, the system includes:

[0052] Semantic analysis module, used to perform semantic analysis on service requests;

[0053] a classification module, classifying the service request based on the result of the semantic analysis;

[0054] a prioritization module, which evaluates service urgency based on the semantic analysis results;

[0055] Matching module: matching the service request with the technical support object;

[0056] Allocation module: allocates the service request to the corresponding technical support object according to the matching result.

[0057] Compared with the prior art, the present invention has at least the following beneficial effects:

[0058] The method described in this paper combines semantic analysis, classification, urgency assessment, and personnel skill matching to enhance the intelligence and refinement of IT service request processing. By assessing the real-time requirements and impact of service requests, it rationally adjusts processing priorities, helping the system respond more quickly to important requests. By matching the skills of technical support recipients with historical service data, it enhances the targeted and adaptable nature of service allocation, optimizes resource utilization, improves processing efficiency and service quality, and strengthens the system's ability to handle complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a simplified flowchart of an AI-based intelligent IT service request classification and allocation method in the present invention. DETAILED DESCRIPTION

[0060] The following description of the AI-based intelligent IT service request classification and allocation method and system of the present invention is combined with schematic diagrams, which illustrate preferred embodiments of the present invention. It should be understood that those skilled in the art may modify the present invention described herein while still achieving the beneficial effects of the present invention. Therefore, the following description should be understood as a general guideline for those skilled in the art and not as a limitation of the present invention.

[0061] The present invention is described in more detail in the following paragraphs by way of example with reference to the accompanying drawings. The advantages and features of the present invention will become more apparent from the following description and claims. It should be noted that the drawings are greatly simplified and not to exact scale, and are provided solely for the purpose of assisting in the description of the embodiments of the present invention.

[0062] Example 1

[0063] Please refer to Figure 1 This embodiment discloses an AI (artificial intelligence)-based intelligent IT (information technology) service request classification and allocation method, the method comprising:

[0064] S1. Perform semantic analysis on the service request;

[0065] S2 classifies the service request based on the results of the semantic analysis;

[0066] S3 based on the semantic analysis results of the service request service urgency assessment;

[0067] S4. Match the service request with the technical support object;

[0068] S5. Allocate the service request to a corresponding technical support object based on the matching result and the evaluation result of the service urgency.

[0069] The method disclosed in this embodiment combines semantic analysis, classification, urgency assessment, and personnel skills matching to enhance the intelligence and refinement of IT service request processing. By assessing the real-time requirements and impact of service requests, it rationally adjusts processing priorities, helping the system respond more quickly to important requests. By matching the skills of technical support recipients with historical service data, it enhances the targeted and adaptable nature of service allocation, optimizes resource utilization, improves processing efficiency and service quality, and strengthens the system's ability to handle complex scenarios.

[0070] In this embodiment, the content in the service request includes, but is not limited to, request type, problem description, user information, device information used by the user, and the user's expectations and requirements for problem solving, etc. The service request can be multi-modal input, which means that the user can submit problems in various forms such as text, voice, picture, video, etc. For example, the user can describe the problem by voice ("My computer won't boot up"), or upload a picture of the computer's black screen. The system will automatically recognize the voice content and extract the key information in the picture, and conduct semantic analysis by combining multi-modal data to improve the accuracy of problem understanding and response efficiency.

[0071] It can be understood that the technical support object described in this embodiment can be an enterprise, an institution, a platform operator, a technical support team or the system itself, specifically including but not limited to IT operation and maintenance personnel, customer service personnel, artificial intelligence service systems or automated processing platforms, etc., which are responsible for service provision and processing according to different application scenarios.

[0072] In step S1, the semantic analysis of the service request includes:

[0073] S11. After converting the service request into a request text, perform text preprocessing on the request text;

[0074] S12. Extract key feature vectors from the preprocessed request text or the knowledge base.

[0075] In this embodiment, the text preprocessing includes at least one of text cleaning, text tokenization,词性标注 (positional tagging), and stop word filtering.

[0076] Specifically, text cleaning refers to removing noise data in the text, such as removing HTML (HyperText Markup Language) tags, garbled characters, special symbols, etc., to improve the purity and consistency of the text content. Text tokenization refers to splitting continuous text into individual words or phrases for subsequent semantic analysis and feature extraction. Positional tagging refers to tagging the corresponding word type for each word after tokenization, such as tagging as noun, verb, adjective, etc., to assist in understanding the grammatical structure and semantic information of the text. Stop word filtering refers to removing common but meaningless words from the text, such as "de" (的), "shi" (是), and "zai" (在), etc., to reduce noise interference and improve the efficiency of key information extraction. These preprocessing steps together lay the foundation for feature extraction and subsequent analysis, thereby improving the intelligence level of the system.

[0077] It should be noted that the term "词性标注" in Chinese is directly translated as "positional tagging" here, but it might be more accurately translated as "part-of-speech tagging" in a more standard context. However, following the instruction to preserve the text as accurately as possible, this translation is provided.Furthermore, in this embodiment, extracting key feature vectors includes at least one of the following steps: extracting words from the request text and obtaining a word vector set; extracting keywords from the request text and obtaining a keyword vector set; aggregating the word vectors and the keyword vectors into a grammatical structure vector based on the grammatical structure input by the user; aggregating the word vectors, the keyword vectors and the grammatical structure vectors into a text vector based on the text semantics input by the user.

[0078] In this embodiment, by obtaining the basic word semantics, extracting the core information of the text, integrating the grammatical structure relationship, and finally fusing the information of each level into a unified text representation, the method performs well in processing long texts, polysemous words and implicit semantics, providing a richer and more accurate text representation basis for downstream applications. At the same time, the design supports flexible configuration, and different processing methods can be selected according to specific application scenarios and computing resource limitations. For example, in a resource-constrained environment, only extracting word vectors can provide basic semantic understanding capabilities; in scenarios where processing speed is pursued, extracting keyword vectors can quickly capture text topics; and in specific field applications, extracting syntactic vectors can accurately grasp grammatical structure relationships. This method not only ensures comprehensive performance, but also takes into account the efficiency and flexibility requirements in practical applications.

[0079] The method for extracting word vectors includes but is not limited to the implementation of algorithm models such as Word2Vec (word to vector model), GloVe (global vector word) and FastText (fast text); the method for extracting keyword vectors includes but is not limited to the implementation of algorithm models such as NER (named entity recognition) and BERT-NER (named entity recognition method based on bidirectional encoder representation converter); the method for extracting syntactic vectors includes but is not limited to the implementation of algorithm models such as Transition-based parsing (syntactic analysis based on transfer), Neural dependency parsing (neural network dependency analysis) and Graph-based parsing (graph-based analysis); the method for extracting text vectors includes but is not limited to the implementation of algorithm models such as BERT (bidirectional encoder representation converter), RoBERTa (robust optimization BERT pre-training method) and XLNet (generalized autoregressive pre-training method). Those skilled in the art can select different algorithm models for feature extraction according to actual application requirements. Preferably, at least the BERT (bidirectional encoder representation converter) model is used in the feature extraction process because the BERT model is more suitable for text classification and intent recognition tasks.

[0080] In a specific embodiment, when the BERT algorithm is used for feature extraction, after the text preprocessing is completed, feature extraction is performed in sequence through the word embedding layer, the position embedding layer and the segment embedding layer. Specifically, the word embedding layer maps each word element (token) to a high-dimensional vector space representation of a fixed dimension, capturing the basic semantic features of the word. The position embedding layer assigns sequence position information to each word element, enabling the model to perceive the relative position relationship of the word in the sentence. The segment embedding layer marks the attribution of different text fragments, especially distinguishing sentence boundaries when processing multi-sentence input. These three embeddings are fused into a unified input representation through element-level addition, and then sent to a multi-layer bidirectional Transformer encoder for deep processing. After deep processing by the multi-layer Transformer encoder, BERT can capture the contextual information in the text and generate a context-related vector representation for each word. The final generated context-sensitive vector not only encodes the semantic information of the word itself, but also contains position information, syntactic structure and long-distance dependencies, providing a comprehensive and accurate text feature representation for downstream tasks.

[0081] Furthermore, the method disclosed in this embodiment can also achieve context matching.

[0082] Specifically, the context matching method includes: finding and obtaining historical problems associated with the user's current problem; finding and obtaining historical user information and historical device information associated with the historical information and user information mentioned in the request text; and finding and obtaining at least one of the fault type and fault solution associated with the user's current problem.

[0083] In this embodiment, the above information will be converted into key feature vectors. These key feature vectors do not directly participate in the classification decision, but are used to assist in the semantic analysis of the current problem, match similar historical cases, evaluate the urgency of the problem, or optimize the priority of problem handling. For example, the system can perform similarity comparison in the pre-trained model based on contextual features, filter out problem scenarios similar to the current problem, and recommend more targeted processing strategies or response content. The addition of contextual information enables the system to understand the continuity of user intentions and the background of the problem, thereby improving the accuracy of semantic understanding and the relevance of the response.

[0084] In a specific embodiment, searching and obtaining historical problems associated with the user's current problem refers to searching for technical terms and problem description words repeatedly used by the user in historical requests, such as keywords such as "blue screen", "no response" and "connection failure", problem domain information such as "hardware failure", "software compatibility" and "network connection" mentioned by the user, emotional information such as urgency, anxiety level or satisfaction expressed by the user, time pattern information of problem occurrence, whether the user has repeatedly submitted similar questions in the past, etc. All of this information is converted into key feature vectors by the first context comparison unit.

[0085] In another specific embodiment, when searching and obtaining historical user information and historical device information associated with the historical information and user information mentioned in the request text, the scattered user information (such as department, position, authority level) and device information (such as operating system, software version, hardware configuration) are first organized into an organized structured format for easy computer processing.

[0086] In another specific embodiment, when searching and obtaining the fault type and fault solution associated with the user's current problem, a query is performed based on a knowledge graph. The knowledge graph integrates the collective knowledge of users, experts, and documents to build a universal knowledge network, representing the organization's overall IT service knowledge base, and can provide standardized solution support for various users.

[0087] It can be understood that the knowledge graph integrates the collective knowledge of all users, experts and documents to build a universal knowledge network, representing the organization's overall IT service knowledge base, and is used to provide standardized solutions for all users.

[0088] Through this approach, combined with historical user request patterns, device information, and specialized solutions from the knowledge graph, multi-dimensional information is converted into key feature vectors and fed into the classification process. This approach, based on a comprehensive understanding of context, improves the accuracy of problem classification and solution matching, ensuring more efficient and accurate subsequent processing.

[0089] Furthermore, in step S2, classifying the service request based on the result of the semantic analysis includes:

[0090] S21. Mapping the key feature vector to the category probability distribution of the service request;

[0091] S22. Obtain the category of the service request according to the category probability distribution.

[0092] In one specific embodiment, in step S21, the probability of each category represents the correlation between that category and the input service request. For example, in an IT service request scenario, requests can be divided into categories such as "hardware failure," "software issue," and "network connection," providing clear guidance for subsequent processing.

[0093] In the specific implementation process, the calculation of the category probability distribution can adopt at least one classification method selected from Softmax (soft maximum function), SVM (support vector machine), Random Forest (random forest), or KNN (nearest neighbor algorithm). Preferably, the Softmax classification method is adopted, which can directly output the probability distribution of each category in the multi-classification task, facilitating subsequent decision-making and analysis, and can be naturally connected with the vector representation to achieve end-to-end training and reasoning.

[0094] The specific form of the Softmax function is:

[0095]

[0096] Where P(y=j|x) represents the probability that the input vector x belongs to category j, x is the input vector, w j is the weight vector of category j, K is the total number of categories, Represents the input vector x and the category j weight vector w j The dot product of .

[0097] Furthermore, in step S22, based on the obtained category probability distribution, a decision step is executed to determine the final category to which the service request belongs. This decision can be made using a variety of computing resources, such as CPUs, GPUs, FPGAs (field programmable gate arrays), or ASICs (application-specific integrated circuits). Preferably, GPUs or FPGAs can be used to achieve efficient probability distribution analysis and real-time category determination, meeting the high concurrency and high performance requirements of large-scale service request scenarios.

[0098] The above method enables efficient classification and rapid decision-making for service requests. This method calculates the correlation between input text and various classification categories, generates an accurate category probability distribution, and quickly determines the specific category of the service request based on this distribution, ensuring efficient responses. The overall process not only improves classification and decision-making accuracy, but also significantly increases response speed and processing capabilities. It can handle the high concurrency demands of large-scale service request scenarios, provides clear guidance for subsequent service processing, and ultimately optimizes business processes and improves efficiency.

[0099] Furthermore, in this embodiment, performing service urgency assessment based on the semantic analysis result includes:

[0100] S31. Evaluate the service urgency of the service request according to the preset priority judgment rules.

[0101] In a specific embodiment, based on a preset weighted scoring model, scores are calculated for the real-time requirements, processing complexity, and impact on service processing results of the service request. The weight coefficients of each score are combined to calculate the comprehensive priority score of the service request. The priority scoring formula is:

[0102] P = w1·R+w2·C+w3·I;

[0103] Among them, P is the comprehensive priority score, R is the real-time score, C is the processing complexity score, I is the impact score, and w1, w2, and w3 are the weight coefficients of each score.

[0104] In addition, the method also includes: sorting service requests according to the comprehensive priority score, giving priority to scheduling high-priority service requests into the high-speed processing channel, and low-priority requests into the buffer queue and executing them in sequence, thereby improving the real-time processing of key tasks and the overall processing efficiency of the system.

[0105] Furthermore, in step S4, matching the service request with the technical support object includes:

[0106] S41. Calculate the similarity between the request text and the technical support object skill description text to obtain a content matching score;

[0107] S42. Calculate the correlation between the service request and the technical support object to obtain a collaborative filtering score;

[0108] S43. Perform weighted fusion of the content matching score and the collaborative filtering score to generate a comprehensive matching score;

[0109] S44. Match the technical support object with the service request based on the comprehensive matching score.

[0110] In a specific embodiment of S41, the similarity is calculated as follows:

[0111] similarity(A,B)=(A·B) / (||A||·|B||);

[0112] Among them, A and B are texts represented by vectors; ||A|| and ||B|| are the respective vector norms.

[0113] It can be understood that cosine similarity is an indicator that measures the degree of similarity between the directions of two vectors. Its value range is between -1 and 1. The closer the value is to 1, the more similar the directions of the two vectors are. The closer the value is to -1, the more opposite the directions of the two vectors are. The value of 0 indicates that the two vectors are orthogonal (i.e., no correlation).

[0114] In a specific embodiment of S42, based on historical work order data, the historical work order processing patterns of similar users are analyzed, and the correlation between service requests and technical support objects is calculated to obtain a collaborative filtering score. The collaborative filtering score is calculated as follows:

[0115]

[0116] Among them, similarity(i,j) represents the collaborative filtering score, R i and R j and represent the ratings or interaction behavior vectors of user i and user j in historical tickets, respectively. The calculation results are used to measure the similarity between users to assist in recommending technical support candidates.

[0117] In a specific embodiment, the interactive behavior vector is a vectorized representation of behavioral features such as clicks, resolution time, and feedback type.

[0118] In a step of S44 , the technical support object having the highest score after weighted fusion of the collaborative filtering score and the similarity score and currently having processing capability and available status is preferentially selected, and the service request is assigned to it for processing.

[0119] Furthermore, the method further includes optimizing and updating the knowledge base, as follows:

[0120] S51. Calculate the TF-IDF (term frequency - inverse document frequency) value of each word in the request text;

[0121] S52 extracts subject information from the request text;

[0122] S53 selects key information in the request text based on the size of the TF-IDF value and the relevance between the subject information;

[0123] S54. Store the key information.

[0124] This embodiment integrates a dual mechanism of TF-IDF statistical analysis and topic screening. A calculation unit objectively assesses word importance, a topic screening unit extracts core text topics, and a knowledge base construction unit performs screening based on word importance and topic relevance. This combined statistical and semantic approach captures document-specific keywords while retaining topic-related information, overcoming the limitations of a single approach and enabling more comprehensive and accurate extraction of key information.

[0125] In a specific embodiment of S51, the TF-IDF value is calculated as follows:

[0126] TFIDF(t,d)=TF(t,d)·IDF(t);

[0127] Where TF(t,d) is the term frequency of word t in document d, and IDF(t) is the inverse document frequency of word t. A higher TF-IDF value indicates that the word appears frequently in the current document and may be a keyword for the current service request.

[0128] In a specific embodiment of S52, topics are extracted from the text based on LDA (Latent Dirichlet Allocation), HDP (Hierarchical Dirichlet Process Model), and BTM (Two-Word Topic Model). These models typically use multiple strategies to determine topic relevance: first, the probability distribution similarity between the text and each topic is calculated, and the topic with the highest probability is selected; second, metrics such as cosine similarity or KL divergence are used to compare the proximity between the text vector and the topic vector; another method is to use the overlap rate between the topic word and the text keyword to make a judgment. In addition, a similarity threshold can be set to retain only topics that exceed the threshold, or a hierarchical clustering method can be used to classify semantically similar topics, thereby achieving topic matching and screening.

[0129] In a specific embodiment of step S53, only words that simultaneously satisfy the TF-IDF value higher than the statistical threshold and the topic relevance higher than the semantic threshold are determined as key information. Specifically, the key information includes but is not limited to the core word itself, the context information of the core word, and the topic identifier.

[0130] Finally, in a specific embodiment of step S54, a knowledge graph or a relational database is used to store these key information, and an association relationship network between words and between words and topics is established, so that the knowledge base not only stores key information, but also retains the semantic connection between information. The knowledge base provides a basis for subsequent knowledge reasoning and application.

[0131] In this embodiment, once this key information is stored in the knowledge base, it serves as the foundational data source for subsequent semantic matching and processing strategy reasoning. Specifically, based on the extracted keywords and topic information, the system searches the knowledge base for historical cases most similar to the current request. Furthermore, the system calculates a comprehensive responsiveness score for each candidate technical support agent, taking into account the urgency of the service request. Ultimately, based on this comprehensive score, the optimal technical support agent currently available is selected and the service request is assigned to that agent.

[0132] In this example, users can choose a cloud-based SaaS (Software as a Service) model for rapid deployment without the need for in-house infrastructure. Alternatively, they can adopt an on-premises deployment solution to ensure data security and privacy and meet specific compliance requirements. Furthermore, the system provides standardized APIs for seamless integration with existing business systems, supporting customized development and functional expansion. Each of these three implementation methods offers advantages, allowing enterprises to flexibly choose based on their scale, security needs, and IT strategies.

[0133] Preferably, this embodiment further includes a monitoring and feedback optimization step, and the specific method includes:

[0134] Monitor the system operation status in real time, track various performance indicators in the process of knowledge extraction, topic modeling and knowledge base update; collect user feedback during use, including but not limited to user evaluation of knowledge recommendation results, problems encountered during use and improvement suggestions; based on the collected operation data and user feedback, regularly or in real time analyze the system performance and knowledge base quality, and identify deficiencies in the model or knowledge base; based on the analysis results, continuously optimize the topic extraction model, feature extraction strategy and knowledge base content to improve the accuracy and practicality of the system, and ensure that the knowledge base remains effective and advanced in the long term.

[0135] Example 2

[0136] Based on the same inventive concept, this embodiment provides an AI-based intelligent IT service request classification and allocation system, which implements the AI-based intelligent IT service request classification and allocation method disclosed in the first embodiment. The system includes:

[0137] The semantic analysis module is used to perform semantic analysis on service requests.

[0138] A classification module is configured to classify the service request based on the result of the semantic analysis.

[0139] The priority sorting module evaluates the service urgency based on the semantic analysis result.

[0140] Matching module: matches the service request with the technical support object.

[0141] Allocation module: allocates the service request to the corresponding technical support object according to the matching result.

[0142] It can be understood that the technical effects brought about by the AI-based intelligent IT service request classification and allocation system disclosed in this embodiment are consistent with the technical effects brought about by the AI-based intelligent IT service request classification and allocation method disclosed in Example 1, and will not be repeated here.

[0143] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. An AI-based intelligent IT service request classification and allocation method, characterized by: The method comprises: Perform semantic analysis on service requests; classifying the service request based on the result of the semantic analysis; Performing a service urgency assessment on the service request based on the semantic analysis result; Matching the service request with a technical support object; The service request is assigned to a corresponding technical support object according to the matching result and the evaluation result of the service urgency.

2. The AI-based intelligent IT service request classification and allocation method according to claim 1, characterized in that: Semantic analysis of service requests includes: After converting the service request into a request text, performing text preprocessing on the request text; Extract key feature vectors from the preprocessed request text or knowledge base.

3. The AI-based intelligent IT service request classification and allocation method according to claim 2, characterized in that: Extracting key feature vectors from the pre-processed request text or knowledge base includes at least one of the following steps: Extracting words from the request text and obtaining a word vector set; Extracting keywords from the request text and obtaining a keyword vector set; Aggregating the word vector and the keyword vector into a grammatical structure vector according to the grammatical structure input by the user; According to the semantics of the text input by the user, the word vector, the keyword vector and the grammatical structure vector are aggregated into a text vector.

4. The AI-based intelligent IT service request classification and allocation method according to claim 2, characterized in that: Extracting key feature vectors from the pre-processed request text or knowledge base also includes: Find and retrieve historical questions related to the user's current question; Searching and obtaining historical user information and historical device information associated with the historical information and user information mentioned in the request text; Find and obtain the fault type and fault solution associated with the current problem of the user.

5. The AI-based intelligent IT service request classification and allocation method according to claim 2, characterized in that: The text preprocessing includes at least one of text cleaning, text segmentation, part-of-speech tagging, and stop word filtering.

6. The AI-based intelligent IT service request classification and allocation method according to claim 2, characterized in that: The service urgency assessment based on the semantic analysis results includes: According to a preset priority judgment rule, the service request is evaluated for service urgency based on the key feature vector; The priority judgment rules are: P = w1·R+w2·C+w3·I; Among them, P is the comprehensive priority score, R is the real-time score, C is the processing complexity score, I is the impact score, and w1, w2, and w3 are the weight coefficients of each score.

7. The AI-based intelligent IT service request classification and allocation method according to claim 2, characterized in that: Classifying the service request based on the result of the semantic analysis includes: Mapping the key feature vector to a category probability distribution of the service request; Obtaining a category of the service request according to the category probability distribution; The calculation method of the category probability distribution is: Where P(y=j|x) represents the probability that the input vector x belongs to category j, x is the input vector, w j is the weight vector of category j, K is the total number of categories, Represents the input vector x and the category j weight vector w j The dot product of .

8. The AI-based intelligent IT service request classification and allocation method according to claim 1, characterized in that: Allocating the service request to a corresponding technical support object according to the matching result and the evaluation result of the service urgency includes: Calculating the similarity between the request text and the technical support object skill description text to obtain a content matching score; The similarity is calculated as follows: similarity(A,B)=(A·B) / (||A||·|B||); Where A and B are texts represented by vectors; ||A|| and ||B|| are the vector norms of each; Calculate the correlation between service requests and technical support objects to obtain collaborative filtering scores; The collaborative filtering score is calculated as follows: Among them, similarity(i,j) represents the collaborative filtering score, R i and R j Represent the ratings or interaction behavior vectors of user i and user j in historical tickets respectively; Performing weighted fusion on the content matching score and the collaborative filtering score to generate a comprehensive matching score; The technical support object is matched with the service request according to the comprehensive matching score.

9. The AI-based intelligent IT service request classification and allocation method according to claim 1, characterized in that: The method further comprises: Calculate the TF-IDF value of each word in the request text; extracting subject information from the request text; Selecting key information in the request text based on the TF-IDF value and the relevance between the key information and the topic information; and storing the key information; The TF-IDF value is calculated as follows: TF-IDF(t,d)=TF(t,d)·IDF(t); Among them, TF(t,d) is the term frequency of word t in document d, and IDF(t) is the inverse document frequency of word t.

10. An AI-based intelligent IT service request classification and allocation system, characterized by: The system comprises: Semantic analysis module, used to perform semantic analysis on service requests; a classification module, classifying the service request based on the result of the semantic analysis; a prioritization module, which evaluates service urgency based on the semantic analysis results; Matching module: matching the service request with the technical support object; Allocation module: allocates the service request to the corresponding technical support object according to the matching result.

Images