SaaS platform-oriented multi-tenant question collection and automatic classification processing device

Abstract

The invention relates to the field of data management, and discloses a SaaS platform-oriented multi-tenant problem collection and automatic classification processing device, which is used for solving the problem that a multi-tenant collection device for classifying information is lacked when the user quantity feedback problems are too many. Comprising a data collection module, a preprocessing module, an annotation module, a classifier module, a feedback and updating module and a processing module, and is used for collecting request information data of multi-tenant users, preprocessing the request information data, classifying a group of known data in a manual annotation mode, constructing a problem classification model, and analyzing the problem classification model. The problem classification model is used for classifying and marking input unclassified driver problems, classification results are output, parameters of a classifier are updated according to differences between output results and real labels, and whether automatic processing or manual processing is carried out is judged according to problem categories and keyword features in the problems; and the user requirements and the problem solving efficiency are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of data management, and more specifically to a device for collecting and automatically classifying and processing multi-tenant problems on a SaaS platform. Background Art

[0002] With the rapid development of online car-hailing SaaS business, the number of drivers per tenant has increased rapidly. The driver-side App for each tenant of the online car-hailing business is different. As the number of users increases, the problem of the driver-side App being difficult to use is exposed. Users have no feedback channel, and there is a lack of a collection device for multiple tenants to collect this information for classification and processing.

[0003] For these problems, effective classification and processing are needed to better understand user needs and solve problems. However, manual classification and processing of these problems is time-consuming, labor-intensive, and prone to errors. Therefore, a device that can automatically classify and process crowd-testing problems is needed.

[0004] In view of the above problems, the present invention proposes a solution. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a device for collecting and automatically classifying and processing multi-tenant problems on a SaaS platform to solve the problems existing in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A device for collecting and automatically classifying multi-tenant problems on a SaaS platform includes a data collection module, a preprocessing module, a labeling module, a classifier module, a feedback and update module, and a processing module. The modules are connected by signals, and the steps of data processing between the modules are as follows:

[0008] The data collection module is used to collect the request information data of the multi-tenant user APP through the http interface, store the collected data in the database, and transmit the request information data to the pre-processing module;

[0009] A preprocessing module is used to receive the request information data transmitted by the data collection module, preprocess the request information data, discard invalid information, and transmit the preprocessed request information data to the marking module;

[0010] The labeling module is used to receive the preprocessed request information data transmitted by the preprocessing module, supervise the reference system part in the learning system, classify a set of known data through manual labeling, and transmit the labeled data to the classifier module;

[0011] The classifier module is used to receive the labeled data transmitted by the labeling module, build a question classification model based on the labeled data, and use the question classification model to classify and label the input unclassified driver questions, that is, to classify the input questions into different categories and labels according to their characteristics, and output the classification results;

[0012] Feedback and update module, used to provide feedback information and update the parameters of the classifier according to the difference between the output result of the classifier and the true label;

[0013] The processing module is used to determine whether to automatically or manually process the classified questions based on the question category and keyword features in the questions.

[0014] Preferably, the preprocessing includes text cleaning, text standardization, word segmentation, and stop word removal;

[0015] The text cleaning is to remove irrelevant information, advertisements and links to irrelevant content;

[0016] The text standardization is to convert the text into a unified format;

[0017] The word segmentation is to use the jieba Chinese word segmentation tool to split the text into independent words;

[0018] The stop word removal is to remove common stop words in the text.

[0019] Preferably, the step of classifying a set of known data by manual labeling is:

[0020] Formulate labeling rules and classification standards;

[0021] Classify and label data according to established rules and standards;

[0022] Perform quality check on the annotation results and remove erroneous annotation data.

[0023] Preferably, the step of constructing a question classification model based on the labeled data is:

[0024] Step 1: Based on the results of the preprocessing module, the word2vec embedding method is used to convert the text into a numerical vector and extract features related to the classification task;

[0025] Step 2: Put the labeled data in the labeling module into the dataset and divide the dataset into training set, validation set and test set;

[0026] Step 3: Use the training set to train the model, use the validation set to adjust the parameters, obtain the pre-model, monitor the loss and evaluation indicators during the training process, and adjust the hyperparameters;

[0027] Step 4: Use the test set to evaluate the performance of the pre-model and calculate the performance indicators of the pre-model, which include accuracy, precision, recall and F1 value. The performance index of the model is obtained according to the performance indicator evaluation of the pre-model, and its expression is XN = f(Z 1 ,Z 2 ,Z 3 ,Z 4 ), where XN represents the performance index, Z 1 , Z 2 , Z 3 , Z 4 Expressed as accuracy, precision, recall and F1 value, the performance index is compared with the preset threshold. If the performance index is greater than the preset threshold, the current model is output as the problem classification model. If the performance index is less than the preset threshold, return to step 3 to continue optimizing the model parameters until the performance index is greater than the preset threshold.

[0028] Preferably, the step of determining whether to perform automatic processing or manual processing according to the question category and the keyword features in the question is:

[0029] List all problem categories;

[0030] For each question category, define a set of keywords;

[0031] Extract keyword information from the question, match the category and keyword information with the data in the automatic processing set, and determine how to process it based on the matching results.

[0032] Preferably, the step of extracting keyword information from the question is:

[0033] Use the word segmentation tool to divide the question text into words, remove common but meaningless words, and restore the words to stem form;

[0034] Calculate the frequency of each word in the question text. The calculation expression is: Where TF(t,d) is the word frequency in the question text, t is the word, d is the question text, count(t,d) is the number of times word t appears in text d, and total_words(d) is the total number of words in text d;

[0035] Calculate the inverse document frequency of each word in the entire question set. The calculation expression is: Where IDF(t,D) represents the inverse document frequency of a word in the entire question set, |D| represents the total number of documents in the question set, and |{d∈D:t∈d} represents the number of documents containing word t;

[0036] Combine the word frequency and inverse document frequency to calculate the TF-IDF value of each word, which is used to measure the importance of the word in the question. The calculation formula is TF-IDF(t,d,D)=TF(t,d)×IDF(t,D), where TF-IDF(t,d,D) represents the TF-IDF value of the word, TF(t,d) is the word frequency in the question text, and IDF(t,D) represents the inverse document frequency of the word in the entire question set.

[0037] Arrange the TF-IDF value of each word from high to low, and select the top-ranked words as keywords.

[0038] Preferably, the steps of matching the category and keyword information with the data in the automatic processing set and determining how to process according to the matching result are:

[0039] Use the bag-of-words model and TF-IDF to represent question text and category descriptions as vectors;

[0040] The cosine similarity method is used to calculate the similarity between the customer question and each question category description. The calculation formula is: Where cosine_similarity(A,B) represents similarity, A and B represent vector representations of question and category descriptions, A·B represents the dot product of A and B, and ||A|| represents the modulus of vector A;

[0041] Set a similarity threshold and compare the similarity with the similarity threshold. If the similarity is higher than the threshold, the problem will be automatically processed. If the similarity is lower than the threshold, it will be handled manually.

[0042] Technical effects and advantages of the present invention:

[0043] Collect request information data from multi-tenant users, pre-process the request information data, classify a set of known data through manual labeling, build a problem classification model, use the problem classification model to classify and label the input unclassified driver questions, and output the classification results. The difference between the output results and the true labels is updated, and the parameters of the classifier are updated. According to the problem category and the keyword features in the problem, it is determined whether to perform automatic or manual processing, which effectively improves the efficiency of solving user needs and problems.

[0044] By automating and optimizing existing workflows, the present invention can significantly improve work efficiency and reduce the consumption of manpower and time. By reducing the consumption of manpower and time and optimizing resource utilization, the present invention can help enterprises reduce operating costs. Through automation and intelligent design, the present invention reduces the need for manual intervention and reduces the time consumption of manual operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is the overall flow chart of the present invention. DETAILED DESCRIPTION

[0046] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention. In addition, the forms of the various structures recorded in the following implementation modes are only examples. The present invention involves a multi-tenant problem collection and automatic classification processing device for a SaaS platform and is not limited to the various structures recorded in the following implementation modes. All other implementation modes obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0047] The present invention provides a device for collecting and automatically classifying multi-tenant problems on a SaaS platform, including a data collection module, a preprocessing module, a labeling module, a classifier module, a feedback and update module, and a processing module. The modules are connected by signals, and the steps of data processing between the modules are as follows:

[0048] The data collection module is used to collect the request information data of multi-tenant user APPs through the http interface, store the collected data in the database, provide a data source for subsequent processing and analysis, and transmit the request information data to the preprocessing module;

[0049] A preprocessing module is used to receive the request information data transmitted by the data collection module, preprocess the request information data, and discard invalid information;

[0050] In this embodiment, it should be specifically noted that the preprocessing includes text cleaning, text standardization, word segmentation, and stop word removal;

[0051] The text cleaning is to remove irrelevant information, advertisements, links and other irrelevant contents;

[0052] The text standardization is to convert the text into a unified format, such as lowercase, removing special characters, etc.;

[0053] The word segmentation is to use a Chinese word segmentation tool such as Jieba to split the text into independent words or phrases for subsequent processing and feature extraction. The Jieba Chinese word segmentation tool is an efficient and accurate Chinese word segmentation tool, which is widely used in natural language processing tasks. It supports three word segmentation modes: precise mode, full mode and search engine mode. It can handle ambiguity and perform part-of-speech tagging. Jieba also allows users to load custom dictionaries to optimize the word segmentation effect. It is known for its ease of use, speed and high accuracy, and is the preferred tool for processing Chinese text word segmentation;

[0054] The stop-word removal is to remove common stop words in the text, such as commonly used words like "de" (of), "he" (and), etc., to improve the accuracy of feature extraction.

[0055] The annotation module is used for the reference system part in the supervised learning system. It classifies a set of known data through manual annotation to provide a basis for setting the parameters of the classifier.

[0056] Manual annotation can create a standard data set for training and evaluating machine learning models. This data set contains correct classification information, serving as the learning target of the model and the standard reference for evaluating performance; manual annotation can provide highly accurate and reliable data classification results. Compared with automatic annotation or unannotated data, manual annotation can avoid misclassification and incorrect data labels, ensuring the quality and accuracy of the data; the data annotated manually can be used to train supervised learning models, such as classifiers or regression models. These models learn using known data features and labels and then make predictions or classifications on unseen data.

[0057] In this embodiment, it should be specifically noted that the steps of classifying a set of known data through manual annotation are as follows:

[0058] Formulate clear annotation rules and classification criteria to ensure the consistency and accuracy of annotation.

[0059] Classify and annotate the data according to the specified rules and criteria.

[0060] Conduct quality inspection on the annotation results, remove inaccurate or incorrect annotated data, and ensure the quality of the annotated data.

[0061] The classifier module is used to construct a question classification model based on the annotated data, and use the question classification model to classify or label the unclassified driver questions input, that is, classify the input questions into different categories or labels according to their features, and output the classification results in an appropriate way, such as displaying on a web page, saving to a database, etc. The results can include information such as classification labels, probabilities, etc.

[0062] In this embodiment, it should be specifically noted that the steps of constructing a question classification model based on the annotated data are as follows:

[0063] Step 1: Based on the results of the preprocessing module, extract features related to the classification task. For text classification tasks, text can be converted into numerical vectors using methods such as TF-IDF weighting and word2vec embedding. Word2Vec is a technology for generating word embeddings, which can map words to vector representations in a high-dimensional space. These vectors capture the semantic similarity and grammatical relationship between words, and are therefore widely used in natural language processing tasks such as text classification, information retrieval, and named entity recognition.

[0064] For non-text classification tasks, corresponding features can be extracted according to specific needs.

[0065] Step 2: Put the labeled data in the labeling module into the dataset and divide the dataset into training set, validation set and test set;

[0066] Step 3: Use the training set to train the model, use the validation set to adjust the parameters, obtain the pre-model, monitor the loss and evaluation indicators during the training process, and adjust the hyperparameters to prevent overfitting;

[0067] Step 4: Use the test set to evaluate the performance of the pre-model and calculate the performance indicators of the pre-model, which include accuracy, precision, recall and F1 value. The performance index of the model is obtained according to the performance indicator evaluation of the pre-model, and its expression is XN = f(Z 1 ,Z 2 ,Z 3 ,Z 4 ), where XN represents the performance index, Z 1 , Z 2 , Z 3 , Z 4 Expressed as accuracy, precision, recall and F1 value, the performance index is compared with the preset threshold. If the performance index is greater than the preset threshold, the current model is output as the problem classification model. If the performance index is less than the preset threshold, return to step 3 to continue optimizing the model parameters until the performance index is greater than the preset threshold.

[0068] The accuracy is the ratio of all correctly classified samples to the total number of samples, and its calculation formula is: Where Z 1 is the accuracy rate, TP is the number of samples correctly predicted as positive by the model, TN is the number of samples correctly predicted as negative by the model, FP is the number of samples incorrectly predicted as positive by the model, and FN is the number of samples incorrectly predicted as negative by the model;

[0069] The precision rate is the proportion of positive samples in the positive class that are actually positive, and its calculation formula is: Where Z 2 is the accuracy rate;

[0070] The recall rate is the proportion of samples that are actually positive that are correctly predicted as positive by the model. The calculation formula is: Where Z 3 is the recall rate;

[0071] The F1 value is the harmonic mean of precision and recall, which is used to comprehensively measure the precision and recall of the model. Its calculation formula is: Where Z 4 is the F1 value, Z 2 is the accuracy, Z 3 is the recall rate.

[0072] The training set is used to train the machine learning model. With labeled data, the model can learn the relationship between the features of the data and the labels, thereby improving the prediction or classification ability on unknown data; the validation set is used to evaluate the generalization ability of the model during the training process. The model trained on the training set needs to be evaluated on the validation set to determine whether the model is overfitting or underfitting, and to optimize the parameters; by comparing the performance of different models or different parameter settings of the same model on the validation set, the best performing model can be selected as the final model for practical application or further testing; using different data set partitioning methods (such as cross-validation) can enhance the robustness of the model. This method can reduce the contingency or randomness caused by the partitioning of a specific data set, thereby better evaluating the general performance of the model.

[0073] Feedback and update module, which is used to provide feedback information and update the parameters of the classifier according to the difference between the output of the classifier and the true label, so as to continuously optimize the performance and accuracy of the classifier. This module is usually closely related to the training process and model optimization, and is used to improve the prediction ability and generalization ability of the classifier;

[0074] In this embodiment, it should be specifically explained that the steps of providing feedback information and updating the parameters of the classifier are:

[0075] Real-time monitoring of classifier performance indicators, such as accuracy and recall, to promptly detect possible problems or overfitting of the classifier;

[0076] Adjust and optimize the classifier based on the monitoring results, such as adjusting model parameters, changing algorithms, etc. For deep learning models, it may be necessary to retrain some or all layers to update the model;

[0077] As time goes by and data accumulates, new labeled data is collected and updated to the training set. In order to maintain the real-time and accuracy of the classifier, the training data and model need to be updated regularly;

[0078] In order to further improve the stability and accuracy of classification performance, ensemble learning algorithms are used to fuse multiple classifiers. Ensemble learning algorithms such as bagging and boosting can combine the prediction results of multiple basic classifiers to obtain more reliable classification results.

[0079] The processing module is used to determine whether to automatically or manually process the classified questions based on the question category and keyword features in the questions;

[0080] In this embodiment, it should be specifically explained that the steps of determining whether to perform automatic processing or manual processing according to the question category and the keyword features in the question are:

[0081] List all possible problem categories, such as technical issues, account issues, payment issues, service requests, etc.;

[0082] For each question category, define a set of keywords, for example:

[0083] Account issues: password, reset, login, account lock;

[0084] Payment issues: payment failure, refund, bill, and record the keywords corresponding to each problem category as the automatic processing set;

[0085] Extract keyword information from the question, match the category and keyword information with the data in the automatic processing set, and determine how to process it based on the matching results.

[0086] In this embodiment, it should be specifically explained that the step of extracting keyword information in the question is:

[0087] Use existing word segmentation tools to divide the question text into words or phrases, and remove common but meaningless words;

[0088] Restore words to their stem form to reduce the impact of word form changes on keyword extraction;

[0089] Calculate the frequency of each word in the question text. The calculation expression is: Where TF(t,d) is the word frequency in the question text, t is the word, d is the question text, count(t,d) is the number of times word t appears in text d, and total_words(d) is the total number of words in text d;

[0090] Calculate the inverse document frequency of each word in the entire question set. The calculation expression is: Where IDF(t,D) represents the inverse document frequency of a term in the entire question set, |D| represents the total number of documents in the question set, and |{d∈D:t∈d}| represents the number of documents containing the term t.

[0091] Combine the word frequency and inverse document frequency to calculate the TF-IDF value of each word, which is used to measure the importance of the word in the question. The calculation formula is TF-IDF(t,d,D)=TF(t,d)×IDF(t,D), where TF-IDF(t,d,D) represents the TF-IDF value of the word, TF(t,d) is the word frequency in the question text, and IDF(t,D) represents the inverse document frequency of the word in the entire question set.

[0092] Arrange the TF-IDF value of each word from high to low, and select the top-ranked words as keywords.

[0093] In this embodiment, it should be specifically explained that the steps of matching the category and keyword information with the data in the automatic processing set and determining how to process according to the matching results are as follows:

[0094] The question text and category description are represented as vectors using the bag-of-words model and TF-IDF. The bag-of-words model is a basic method in natural language processing and text mining. It is used to represent text data and represents text as a word frequency vector without considering the order or grammar of the words, but only focusing on the frequency of occurrence of the words. TF-IDF is a statistical method commonly used in text mining and information retrieval to evaluate the importance of a word in a document collection or corpus. It combines word frequency and inverse document frequency to reduce the impact of common but meaningless words and highlight important words.

[0095] The cosine similarity method is used to calculate the similarity between the customer question and each question category description. The calculation formula is: Where cosine_similarity(A,B) represents similarity, A and B represent vector representations of question and category descriptions, A·B represents the dot product of A and B, and ||A|| represents the modulus of vector A. The cosine similarity is a method for measuring the similarity between two vectors, which is commonly used in fields such as information retrieval and text mining. It measures the difference in direction between two vectors, rather than the difference in size or length. In natural language processing, cosine similarity is often used to compare the similarity of texts;

[0096] Set a similarity threshold and compare the similarity with the similarity threshold. If the similarity is higher than the threshold, the problem will be automatically processed. If the similarity is lower than the threshold, it will be handled manually.

[0097] Automatic processing can handle common, standardized problems quickly and instantly without human intervention. This can save a lot of human resources and time, allowing human processing personnel to focus on more complex problems that require human intelligence and judgment; automatic processing can significantly reduce the cost of problem handling. Compared with manual processing, automated processing does not require labor costs and can run 24 / 7 uninterruptedly, thereby improving the efficiency and cost-effectiveness of services; automatic processing can respond to user requests instantly, process and solve problems immediately, and is not restricted by working hours and human resources. This can significantly shorten user waiting time and enhance the immediacy of service and user satisfaction.

[0098] Automatic processing can perform standardized processing according to preset rules and algorithms, avoiding errors and inconsistencies that may be introduced by human factors. This helps to improve the accuracy and consistency of problem handling; classifying problems into automatic processing and manual processing can allocate human resources more accurately. Manual processing can focus on those problems that require professional knowledge, complex judgments or situational understanding, while automatic processing can handle a large number of repetitive problems, making the best use of resources; automatic processing can quickly solve common user problems, provide instant help and feedback, and thus enhance user satisfaction and experience. At the same time, for complex problems, since manual processing resources can be allocated in a targeted manner, more professional and personalized services can also be provided.

[0099] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0100] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A device for collecting and automatically classifying multi-tenant issues on a SaaS platform, characterized in that: It includes data collection module, preprocessing module, labeling module, classifier module, feedback and update module and processing module. The modules are connected by signals. The steps of data processing between modules are as follows: The data collection module is used to collect the request information data of the multi-tenant user APP through the http interface, store the collected data in the database, and transmit the request information data to the pre-processing module; A preprocessing module is used to receive the request information data transmitted by the data collection module, preprocess the request information data, discard invalid information, and transmit the preprocessed request information data to the marking module; The labeling module is used to receive the preprocessed request information data transmitted by the preprocessing module, supervise the reference system part in the learning system, classify a set of known data through manual labeling, and transmit the labeled data to the classifier module; The classifier module is used to receive the labeled data transmitted by the labeling module, build a question classification model based on the labeled data, and use the question classification model to classify and label the input unclassified driver questions, that is, classify the input questions into different categories and labels according to their characteristics, and output the classification results; Feedback and update module, used to provide feedback information and update the parameters of the classifier according to the difference between the output result of the classifier and the true label; The processing module is used to determine whether to automatically or manually process the classified questions based on the question category and keyword features in the questions.

2. According to claim 1, a device for collecting and automatically classifying multi-tenant issues on a SaaS platform, characterized in that: The preprocessing includes text cleaning, text standardization, word segmentation and stop word removal; The text cleaning is to remove irrelevant information, advertisements and links to irrelevant content; The text standardization is to convert the text into a unified format; The word segmentation is to use the jieba Chinese word segmentation tool to split the text into independent words; The stop word removal is to remove common stop words in the text.

3. According to claim 1, a device for collecting and automatically classifying multi-tenant issues on a SaaS platform is characterized in that: The steps of classifying a set of known data by manual labeling are as follows: Formulate labeling rules and classification standards; Classify and label data according to established rules and standards; Perform quality check on the annotation results and remove erroneous annotation data.

4. The device for collecting and automatically classifying and processing multi-tenant problems on a SaaS platform according to claim 1 is characterized in that: The steps of constructing a problem classification model based on the labeled data are as follows: Step 1: Based on the results of the preprocessing module, the word2vec embedding method is used to convert the text into a numerical vector and extract features related to the classification task; Step 2: Put the labeled data in the labeling module into the dataset and divide the dataset into training set, validation set and test set; Step 3: Use the training set to train the model, use the validation set to adjust the parameters, obtain the pre-model, monitor the loss and evaluation indicators during the training process, and adjust the hyperparameters; Step 4: Use the test set to evaluate the performance of the pre-model and calculate the performance indicators of the pre-model, which include accuracy, precision, recall and F1 value. The performance index of the model is obtained according to the performance indicator evaluation of the pre-model, and its expression is XN=f(Z1, Z2, Z3, Z4), where XN represents the performance index, Z1, Z2, Z3, Z4 represent accuracy, precision, recall and F1 value. The performance index is compared with the preset threshold. If the performance index is greater than the preset threshold, the current model is output as the problem classification model. If the performance index is less than the preset threshold, return to step 3 to continue optimizing the model parameters until the performance index is greater than the preset threshold.

5. The device for collecting and automatically classifying and processing multi-tenant problems on a SaaS platform according to claim 1 is characterized in that: The steps of determining whether to perform automatic processing or manual processing according to the question category and the keyword features in the question are: List all problem categories; For each question category, define a set of keywords; Extract keyword information from the question, match the category and keyword information with the data in the automatic processing set, and determine how to process it based on the matching results.

6. The device for collecting and automatically classifying and processing multi-tenant issues on a SaaS platform according to claim 5, characterized in that: The steps of extracting keyword information in the question are: Use the word segmentation tool to divide the question text into words, remove common but meaningless words, and restore the words to stem form; Calculate the frequency of each word in the question text. The calculation expression is: Where TF(t,d) is the word frequency in the question text, t is the word, d is the question text, count(t,d) is the number of times word t appears in text d, and total_words(d) is the total number of words in text d; Calculate the inverse document frequency of each word in the entire question set. The calculation expression is: Where IDF(t,D) represents the inverse document frequency of a term in the entire question set, |D| represents the total number of documents in the question set, and |{d∈D:t∈d}| represents the number of documents containing the term t. Combine the word frequency and inverse document frequency to calculate the TF-IDF value of each word, which is used to measure the importance of the word in the question. The calculation formula is TF-IDF(t,d,D)=TF(t,d)×IDF(t,D), where TF-IDF(t,d,D) represents the TF-IDF value of the word, TF(t,d) is the word frequency in the question text, and IDF(t,D) represents the inverse document frequency of the word in the entire question set. Arrange the TF-IDF value of each word from high to low, and select the top-ranked words as keywords.

7. The device for collecting and automatically classifying and processing multi-tenant issues on a SaaS platform according to claim 5, characterized in that: The steps of matching the category and keyword information with the data in the automatic processing set and determining how to process the data according to the matching results are as follows: Use the bag-of-words model and TF-IDF to represent question text and category descriptions as vectors; The cosine similarity method is used to calculate the similarity between the customer question and each question category description. The calculation formula is: Where cosine_similarity(A,B) represents similarity, A and B represent vector representations of question and category descriptions, A·B represents the dot product of A and B, and ||A|| represents the modulus of vector A; Set a similarity threshold and compare the similarity with the similarity threshold. If the similarity is higher than the threshold, the problem will be automatically processed. If the similarity is lower than the threshold, it will be handled manually.

Images