Person image-based service problem ticket classification method and device

Abstract

The application discloses a business question sheet classification method and device based on a character portrait, relates to the field of artificial intelligence, and comprises the following steps: acquiring a business question sheet, extracting text features from the business question sheet, obtaining label information of a user who proposes the business question sheet, determining character features based on the label information, fusing the character features and the text features to obtain two business question sheet related features, classifying the two business question sheet related features, and judging whether the two business question sheets are the same or not. According to the application, the features of the business question sheet are extracted from multiple aspects according to actual conditions, and the analysis efficiency of the business question sheet can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and apparatus for classifying business problem sheets based on user profiles. Background Technology

[0002] This section is intended to provide background or context for the embodiments of the invention set forth in the claims. The description herein is not an admission that it is prior art simply because it is included in this section.

[0003] For data migration work, business personnel are generating a massive number of issue tickets, overwhelming requirements analysts and developers. However, despite the seemingly large number of these issue tickets, many of them contain duplicate and redundant problems.

[0004] Traditionally, solutions to this type of problem involve using Natural Language Processing (NLP) algorithms to extract features from the question description section of the problem slip, i.e., the text itself, to build a binary classification model to determine whether the question is a duplicate of a previous question. However, because the text portion of problem slips is often short in business scenarios, using only text features to judge business problem slips is not comprehensive and can easily lead to low classification accuracy. Summary of the Invention

[0005] This invention provides a business problem list classification method based on user profiles to improve classification accuracy. The method includes:

[0006] Obtain business issue tickets and extract text features from them;

[0007] Obtain the tag information of the user who submitted the business question, and determine the user's characteristics based on the tag information;

[0008] By fusing character features and text features, relevant features of the two business question sheets can be obtained;

[0009] Classify the relevant features of two business issue tickets to determine whether they are the same business issue ticket.

[0010] This invention also provides a business problem classification device based on a person's profile to improve classification accuracy. The device includes:

[0011] The text feature extraction module is used to obtain business problem slips and extract text features from them.

[0012] The user feature extraction module is used to obtain the tag information of the user who submitted the business question, and to determine the user features based on the tag information;

[0013] The feature fusion module is used to fuse human features and text features to obtain relevant features between the two business question forms;

[0014] The feature classification module is used to classify the relevant features of two business issue orders and determine whether they are the same business issue orders.

[0015] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described business problem list classification method based on person profile.

[0016] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described business problem classification method based on person profiles.

[0017] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described business problem list classification method based on person profile.

[0018] In this embodiment of the invention, compared with the existing technology that only uses text features to judge business problem slips, which is not comprehensive and easily leads to low classification accuracy, this invention obtains business problem slips, extracts text features from them, obtains the tag information of the user who submitted the business problem slip, and determines the person's characteristics based on the tag information; fuses the person's characteristics and text features to obtain the relevant features of two business problem slips; and classifies the relevant features of the two business problem slips to determine whether they are the same business problem slip. This invention performs immersive feature extraction based on the actual situation, focuses on the person profile of the business personnel, and tags them for their work attitude, emotions, personality traits, etc., to obtain person characteristics. By extracting features of business problem slips from multiple levels, this invention can effectively improve the analysis efficiency of business problem slips. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0020] Figure 1 This is the flowchart of the business problem form classification method based on user profile in this embodiment of the invention. Figure 1 ;

[0021] Figure 2 This is the flowchart of the business problem form classification method based on user profile in this embodiment of the invention. Figure 2 ;

[0022] Figure 3 This is the flowchart of the business problem form classification method based on user profile in this embodiment of the invention. Figure 3 ;

[0023] Figure 4 This is the flowchart of the business problem form classification method based on user profile in this embodiment of the invention. Figure 4 ;

[0024] Figure 5 This is the flowchart of the business problem form classification method based on user profile in this embodiment of the invention. Figure 5 ;

[0025] Figure 6 This is a structural block diagram of a business problem classification device based on person profile in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0027] This invention conducts a detailed analysis of business issue requests and finds that: First, business issue requests generated during data migration drills typically have limited text content and are short, making text-based judgments limited in information retrieval and easily leading to low classification accuracy. Second, other factors in data migration drill scenarios can affect the quality of business issue requests. For example, the characteristics of the business personnel submitting the issue requests (key participants such as creators, intermediaries, and processors) are important. Generally, it is believed that responsible business personnel are more likely to submit high-quality issue requests (involving their work attitude). Furthermore, as migration drills progress, technical difficulties can sometimes delay the work, affecting the mood of business personnel and potentially impacting the quality of the issues raised. Therefore, ignoring such scenario-related personnel characteristics can significantly reduce classification accuracy. Additionally, screenshots provided by business personnel and code used for testing also affect the quality of business issue requests.

[0028] Based on this, the present invention integrates the influence of these features on the results and proposes a business problem single classification method based on user profiles, such as... Figure 1 As shown, the method includes:

[0029] Step 101: Obtain business issue tickets and extract text features from them;

[0030] Step 102: Obtain the tag information of the user who submitted the business problem form, and determine the user characteristics based on the tag information;

[0031] Step 103: Perform feature fusion on the personnel features and text features corresponding to each business problem form to obtain a feature vector;

[0032] Step 104: Perform a similarity test on the feature vectors corresponding to the two business problem sheets to obtain the classification results.

[0033] In embodiments of the present invention, such as Figure 2 As shown, extracting image features from the business problem form in step 101 of the present invention may include:

[0034] Step 201: Use the BERT algorithm to embed the text in the business question form to obtain the text feature vector.

[0035] Specifically, each issue form includes a problem description, which is a text segment. The BERT algorithm is used to embed the text in the business issue form to obtain text feature vectors. The specific steps are as follows:

[0036] First, the input text data is segmented into words and formatted into the format required by the BERT model, generating several 128×768 vectors. Next, these vectors pass through a multi-layer transformer model. Each transformer component includes an attention mechanism and a feedforward neural network, allowing the text to fully incorporate the semantics of its context. After being layered and extracted by generating deep transformers, the final text feature vector is obtained.

[0037] This invention also augments text features to enhance semantic expression. Specific data augmentation methods include synonym conversion and random concatenation after word segmentation to obtain a larger dataset. For example, if the problem description is "Table a has no data," "no" can be converted to "none." Or, if the description is "Field b in table a contains garbled characters," the text can be segmented and recombined to become "Field b, Table a, contains garbled characters." This approach increases the amount of data, thereby improving the model training effect.

[0038] In this embodiment of the invention, the main focus is on using user characteristics to model user profiles. A user profile is essentially about tagging users. This invention tags users from three perspectives: statistical tags, rule-based tags, and machine learning-based tags. The features selected for each tag type are as follows:

[0039] Statistical tags: gender, age, job level, number of business issue requests submitted in the most recent round.

[0040] Rule-based tags (rules determined based on user behavior): Repeated business problem tickets in the last two rounds exceed a certain proportion.

[0041] Machine learning is used to mine class labels: judging work attitude and personality traits based on others' evaluations, and judging emotional tendencies based on communication during the exercise process.

[0042] In this embodiment of the invention, step 102, determining the person's characteristics based on the tag information, includes:

[0043] The characteristics of the person are determined after the tag information is preprocessed.

[0044] Specifically, during data preprocessing, for statistical labels, the string-type feature values ​​need to be encoded. For example, gender requires one-hot encoding, resulting in "01" representing "male" and "10" representing "female," etc. This translates to a 2×1 vector. Age is divided into three stages: under 30, 30-50, and over 50, and encoded as a 3×1 vector. Job level is divided into three levels: low, medium, and high, also converted into a 3×1 vector.

[0045] When performing data preprocessing for rule-based tags, a corresponding proportion can be set for each rule-based tag, such as 20%. In this case, "01" can indicate that repeated business problem orders in the last two rounds exceed 20% of the total business problem orders, while "10" indicates that they do not exceed 20%.

[0046] For machine learning to mine class labels, big data technology is used to collect performance evaluation data of relevant employees from platforms such as human resources systems, and similar training datasets are used to build classifiers to train relevant numerical values. For example, work attitude features can be represented by "-1" to indicate a negative attitude, "0" to indicate a neutral attitude, and "1" to indicate a positive attitude, or by using a vector [1, 0, 0] to represent a negative attitude, a vector [0, 1, 0] to represent a neutral attitude, and a vector [0, 0, 1] to represent a positive attitude.

[0047] Finally, this invention sequentially concatenates statistical tags such as gender, age, and job title, number of business questions, repetition of rule-based features, and machine learning features such as attitude, personality, and mood to form a user feature vector, which serves as the user's profile. This is the user profile modeling process.

[0048] In embodiments of the present invention, such as Figure 3 As shown, step 103 will fuse the person features and text features corresponding to each business question to obtain a feature vector, including:

[0049] Step 301: Use a DCN network to fuse the features of the person and the text to obtain the feature vector.

[0050] Specifically, firstly, the characteristics of the three main types of people are expanded and concatenated along the same dimension, and then embedded into the same dimension as the text features. Here, statistical, rule-based, and machine learning features are concatenated horizontally, such as... Figure 4 As shown, after data preprocessing, gender and repetition are 2×1 vectors, while age, job title, number of business questions, attitude, personality, and mood are 3×1 vectors. Here, the gender and repetition features are multiplied by a matrix initialized to 8×2, and the age and other features are multiplied by a matrix initialized to 8×3, resulting in eight 8×1 feature vectors. These vectors are then horizontally concatenated in order to obtain a 64×1 person feature vector. The text feature is a 128×1 vector output by the BERT model; multiplying it by a 64×128 vector yields a 64×1 vector with the same dimensions as the person feature. Finally, for the concatenation of person and text features, this invention uses a DCN network to cross-combine the two features, extracting more layers of information. Specifically, all features are horizontally concatenated at the embedding layer. In the network layer, the DCN network consists of two parallel parts: a cross network and a deep network. In the cross network, the value of each layer is a linear change of the cross-product between the output values ​​of the previous layer and the embedding layer. Deep networks are essentially regular DNN networks, using deeper neural networks to extract more feature information. Finally, the features obtained from the two networks are concatenated to obtain the features relevant to the business problem.

[0051] In embodiments of the present invention, such as Figure 5 As shown, step 104 performs a similarity test on the feature vectors corresponding to the two business question forms to obtain the classification results, including:

[0052] Step 501: Concatenate the feature vectors corresponding to the two business problem orders into a classifier, use the softmax activation function to transform the feature vectors into a 0-1 probability, and determine whether they are the same business problem order based on the 0-1 probability.

[0053] Specifically, for the feature vectors of two DCN networks concatenated into a classifier, the softmax activation function is used to transform the vectors into a 0-1 probability to determine whether they are the same business problem order. If the probability value is greater than 0.5, the two business problem orders are considered to be duplicates.

[0054] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0055] This invention also provides a business problem list classification device based on user profiles, as described in the following embodiments. Since the principle by which this device solves problems is similar to the business problem list classification method based on user profiles, the implementation of this device can refer to the implementation of the business problem list classification method based on user profiles; repeated details will not be elaborated further.

[0056] Figure 6 This is a structural block diagram of a business problem classification device based on person profile in an embodiment of the present invention, such as... Figure 6 As shown, the business problem classification device based on user profiles includes:

[0057] Text feature extraction module 02 is used to obtain business problem slips and extract text features from the business problem slips.

[0058] The user feature extraction module 04 is used to obtain the tag information of the user who submitted the business question form, and to determine the user features based on the tag information;

[0059] Feature fusion module 06 is used to fuse the person features and text features corresponding to each business problem form to obtain a feature vector;

[0060] Feature classification module 08 is used to perform similarity tests on the feature vectors corresponding to two business problem sheets and obtain classification results.

[0061] In this embodiment of the invention, the text feature extraction module is specifically used for:

[0062] The BERT algorithm is used to embed the text in the business question form to obtain the text feature vector.

[0063] In this embodiment of the invention, the tag information includes statistical tags, rule tags, and machine learning mining tags;

[0064] The statistical tags include the user's gender, age, job level, and the number of business issue requests submitted in the most recent round;

[0065] The rule-based tags include those where the proportion of repeated business problem tickets in the last two rounds exceeds a certain level;

[0066] The machine learning-mined tags include users' work attitude, personality traits, and emotional tendencies.

[0067] In this embodiment of the invention, the human feature extraction module is specifically used for:

[0068] Based on the tag information, the characteristics of the person are determined after data preprocessing.

[0069] In this embodiment of the invention, the human feature extraction module is specifically used for:

[0070] Encode statistical tags;

[0071] Set the appropriate ratio, and encode the rule-type tags based on the appropriate ratio;

[0072] Numerical settings are used to encode the class labels discovered through machine learning.

[0073] In this embodiment of the invention, the feature fusion module is specifically used for:

[0074] The DCN network is used to fuse the features of the person and the text to obtain the feature vector.

[0075] In this embodiment of the invention, the feature classification module is specifically used for:

[0076] The feature vectors corresponding to two business problem orders are concatenated into a classifier. The softmax activation function is used to transform the feature vectors into a 0-1 probability. The 0-1 probability is used to determine whether they are the same business problem order.

[0077] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described business problem list classification method based on person profile.

[0078] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described business problem classification method based on person profiles.

[0079] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described business problem list classification method based on person profile.

[0080] In this embodiment of the invention, compared with the existing technology that only uses text features to judge business problem slips, which is not comprehensive and easily leads to low classification accuracy, this invention obtains business problem slips, extracts text features from them, obtains the tag information of the user who submitted the business problem slip, and determines the person's characteristics based on the tag information; fuses the person's characteristics and text features to obtain the relevant features of two business problem slips; classifies the relevant features of the two business problem slips to determine whether they are the same business problem slip. This invention performs immersive feature extraction based on the actual situation, focuses on the person profile of the business personnel, and tags their work attitude, emotions, personality traits, etc. to obtain person characteristics. By extracting features of business problem slips from multiple levels, this invention can effectively improve the analysis efficiency of business problem slips, optimize the overall quality of business problem slips, reduce human and material costs, and improve the communication efficiency between business personnel and developers.

[0081] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0082] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0083] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0084] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0085] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A business problem classification method based on user profiles, characterized in that, include: Obtain business issue tickets and extract text features from them; Obtain the tag information of the user who submitted the business question, and determine the user's characteristics based on the tag information; The feature vector is obtained by fusing the personnel features and text features corresponding to each business problem form. Perform a similarity test on the feature vectors corresponding to the two business problem orders to obtain the classification results; The tag information includes statistical tags, rule tags, and machine learning mining tags; The statistical tags include the user's gender, age, job level, and the number of business issue requests submitted in the most recent round; The rule-based tags include those where the proportion of repeated business problem orders in the last two rounds exceeds a predetermined percentage; The machine learning-mined tags include users' work attitude, personality traits, and emotional tendencies.

2. The character portrait-based business issue classification method according to claim 1, wherein Extract text features from business issue requests, including: The BERT algorithm is used to embed the text in the business question form to obtain the text feature vector.

3. The character portrait-based business ticket classification method of claim 1, wherein, Determining a person's characteristics based on the tag information includes: Based on the tag information, the characteristics of the person are determined after data preprocessing.

4. The method of claim 3, wherein the character image-based service ticket classification method is characterized by, Based on the tag information, data preprocessing is performed to determine the characteristics of the person, including: Encode statistical tags; Set the appropriate ratio, and encode the rule-type tags based on the appropriate ratio; Numerical settings are used to encode the class labels for machine learning mining.

5. The character portrait-based business ticket classification method of claim 1, wherein, The feature vector is obtained by fusing the personnel characteristics and text features corresponding to each business question, including: The DCN network is used to fuse the features of the person and the text to obtain the feature vector.

6. The character portrait-based business ticket classification method of claim 1, wherein, The similarity of the feature vectors corresponding to the two business question forms is tested to obtain the classification results, including: The feature vectors corresponding to two business problem orders are concatenated into a classifier. The softmax activation function is used to transform the feature vectors into a 0-1 probability. The 0-1 probability is used to determine whether they are the same business problem order.

7. A service ticket classification apparatus based on a person image, characterized by, include: The text feature extraction module is used to obtain business problem slips and extract text features from them. The user feature extraction module is used to obtain the tag information of the user who submitted the business question, and to determine the user features based on the tag information; The feature fusion module is used to fuse the personnel features and text features corresponding to each business problem form to obtain a feature vector. The feature classification module is used to perform similarity tests on the feature vectors corresponding to two business question forms and obtain classification results; The tag information includes statistical tags, rule tags, and machine learning mining tags; The statistical tags include the user's gender, age, job level, and the number of business issue requests submitted in the most recent round; The rule-based tags include those where the proportion of repeated business problem orders in the last two rounds exceeds a predetermined percentage; The machine learning-mined tags include users' work attitude, personality traits, and emotional tendencies.

8. The character portrait-based service ticket classification apparatus according to Claim 7, wherein The text feature extraction module is specifically used for: The BERT algorithm is used to embed the text in the business question form to obtain the text feature vector.

9. The character portrait-based service ticket classification apparatus according to Claim 7, wherein The character feature extraction module is specifically used for: Based on the tag information, the characteristics of the person are determined after data preprocessing.

10. The character portrait-based service ticket classification apparatus according to Claim 9, wherein The character feature extraction module is specifically used for: Encode statistical tags; Set the appropriate ratio, and encode the rule-type tags based on the appropriate ratio; Numerical settings are used to encode the class labels discovered through machine learning.

11. The character portrait-based business ticket classification apparatus according to Claim 7, wherein The feature fusion module is specifically used for: The DCN network is used to fuse the features of the person and the text to obtain the feature vector.

12. The character portrait-based service ticket classification apparatus according to Claim 7, wherein The feature classification module is specifically used for: The feature vectors corresponding to two business problem orders are concatenated into a classifier. The softmax activation function is used to transform the feature vectors into a 0-1 probability. The 0-1 probability is used to determine whether they are the same business problem order.

13. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the business problem classification method based on person profile as described in any one of claims 1 to 6.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the business problem classification method based on person profile as described in any one of claims 1 to 6.

15. A computer program product, characterised in that, The computer program product includes a computer program that, when executed by a processor, implements the business problem classification method based on person profile as described in any one of claims 1 to 6.

Images