Intelligent inspection option anomaly identification method and device, and electronic device

Abstract

The application provides a kind of intelligent inspection option exception identification method, device and electronic equipment, the method comprises: obtaining the questionnaire data to be analyzed of intelligent inspection system, the questionnaire data to be analyzed includes at least one by question and answer to be measured question and answer pair;Each question and answer pair to be measured is carried out semantic reasoning, obtains the sentiment tendency of question and answer pair to be measured, sentiment tendency is used to express the sentiment tendency expressed by question and answer pair to be measured is positive or negative;According to the sentiment tendency of each question and answer pair to be measured, determine the analysis result of each question and answer pair to be measured, analysis result is used to express whether the inspection project associated with question exists security risk.The method, device and equipment can be quickly analyzed for different formats, different content of questionnaire data, determine whether the inspection project associated with question in questionnaire data exists security risk, improve the efficiency and accuracy of questionnaire data analysis.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, and electronic device for intelligent inspection option anomaly identification. Background Technology

[0002] Data center maintenance personnel or inspection robots regularly inspect the data centers and provide feedback on various indicators by filling out relevant questionnaires. After generating the inspection questionnaires, they are sent to the central data center. For unified management, it is necessary to collect and summarize the inspection questionnaire data from each data center and to uniformly analyze and control the operation of each data center.

[0003] However, different data centers maintain their own systems based on the different inspection items and indicators set in the questionnaires, resulting in inconsistent questionnaire formats and content. This leads to low efficiency and high cost in analyzing inspection questionnaire data. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to address the above-mentioned shortcomings of the prior art by providing an intelligent inspection option anomaly identification method, device and electronic device to improve the efficiency of analyzing inspection questionnaire data of different formats and contents.

[0005] In a first aspect, the present invention provides a method for identifying anomalies in intelligent inspection options, the method comprising: Acquire the questionnaire data to be analyzed from the intelligent inspection system, wherein the questionnaire data to be analyzed includes at least one question-answer pair consisting of a question and an answer; Semantic reasoning is performed on each of the question-and-answer pairs to obtain the sentiment tendency of the question-and-answer pairs to be tested. The sentiment tendency is used to characterize whether the sentiment expressed by the question-and-answer pairs to be tested is positive or negative. Based on the sentiment tendency of each question-and-answer pair, the analysis result of each question-and-answer pair is determined. The analysis result is used to characterize whether there are safety hazards in the inspection items associated with the question.

[0006] Furthermore, the step of performing semantic reasoning on each of the question-and-answer pairs to obtain the sentiment tendency of the question-and-answer pairs includes: For each question-answer pair to be tested, the answer of the question-answer pair is matched with preset keywords to obtain the matching result; If the matching result indicates that the answer of the question-and-answer pair matches the preset keywords, the sentiment tendency of the question-and-answer pair is determined based on the preset keywords. If the matching result indicates that the answer of the question-and-answer pair does not match the preset keywords, the question-and-answer pair is input into a pre-trained semantic reasoning model to obtain the sentiment tendency of the question-and-answer pair.

[0007] Furthermore, the pre-trained semantic reasoning model includes a first reasoning model and a second reasoning model. The step of inputting the question-and-answer pair to be tested into a pre-trained semantic reasoning model to obtain the sentiment tendency of the question-and-answer pair includes: The answers to the question-and-answer pairs to be tested are input into the first reasoning model to obtain the first reasoning result; If the first reasoning result includes the sentiment tendency of the answer, the sentiment tendency of the answer of the question-and-answer pair to be tested is determined as the sentiment tendency of the question-and-answer pair to be tested. If the first reasoning result does not include the sentiment of the answer, the question-answer pair to be tested is input into the second reasoning model to obtain the sentiment of the question and the logical constraint value of the answer in the question-answer pair to be tested. The logical constraint value of the answer is the preset value corresponding to the explicit definition of the answer in a logical sense. The logical constraint value is 0 or 1. The sentiment of the question-and-answer pair is determined based on the sentiment tendency of the questions and the logical constraints of the answers in the question-and-answer pair to be tested.

[0008] Further, determining the sentiment tendency of the question-and-answer pair based on the sentiment tendency of the questions and the logical constraint values ​​of the answers includes: If the sentiment of the question is positive and the logical constraint value of the answer is 1, or if the sentiment of the question is negative and the logical constraint value of the answer is 0, then the sentiment of the question to be tested is determined to be positive. If the sentiment of the question is negative and the logical constraint value of the answer is 1, or if the sentiment of the question is positive and the logical constraint value of the answer is 0, then the sentiment of the question to be tested is determined to be negative.

[0009] Furthermore, before inputting the question-answer pair into a pre-trained semantic reasoning model to obtain the sentiment tendency of the answers to the question-answer pair, the method further includes: Obtain historical questionnaire data, which includes multiple historical question-answer pairs; The first question-and-answer pair is selected from the multiple historical question-and-answer pairs, and the answers of the first question-and-answer pair have a sentiment bias. The sentiment of each correct answer in the first question is labeled to obtain the first feature information label; Construct a first training sample based on each of the first question-answer pairs and its corresponding first feature information label; Based on multiple first training samples, a preset semantic reasoning model is trained until the training stopping condition is met, and the trained first reasoning model is obtained. A second question-and-answer pair is selected from the multiple historical question-and-answer pairs. The questions in the second question-and-answer pair have an emotional bias, and the answers have logically limiting significance. For each second question-and-answer pair, the sentiment tendency of the question and the logical constraint value of the answer are labeled to obtain the second feature information label; Construct a second training sample based on each second question-answer pair and its corresponding second feature information label; Based on multiple second training samples, a preset semantic reasoning model is trained until the training stopping condition is met, resulting in a trained second reasoning model.

[0010] Further, the step of matching the answers of the question-and-answer pair to be tested with preset keywords to obtain matching results includes: The answers to the question-and-answer pairs to be tested are matched with preset keywords using regular expressions to obtain the matching results.

[0011] Furthermore, determining the analysis result for each question-and-answer pair based on its sentiment tendency includes: If the sentiment of the question-and-answer pair to be tested is positive, it is determined that there are no safety hazards in the inspection items associated with the questions in the question-and-answer pair to be tested. If the sentiment of the question-and-answer pair to be tested is negative, it is determined that the inspection item associated with the question in the question-and-answer pair has a safety hazard.

[0012] In a second aspect, the present invention provides a questionnaire data processing apparatus, comprising: The first acquisition module is used to acquire the questionnaire data to be analyzed from the intelligent inspection system. The questionnaire data to be analyzed includes at least one question-answer pair consisting of questions and answers. The reasoning module is used to perform semantic reasoning on each of the question-and-answer pairs to obtain the sentiment tendency of the question-and-answer pairs to be tested. The sentiment tendency is used to characterize whether the sentiment expressed by the question-and-answer pairs to be tested is positive or negative. The first determining module is used to determine the analysis result of each question-and-answer pair based on the sentiment tendency of each question-and-answer pair to be tested. The analysis result is used to characterize whether there are safety hazards in the inspection items associated with the question.

[0013] Thirdly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the intelligent inspection option anomaly identification method described in the first aspect.

[0014] Fourthly, the present invention provides an electronic device, including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the intelligent inspection option anomaly identification method as described in the first aspect.

[0015] The intelligent inspection option anomaly identification method, device, and electronic device provided by this invention first acquires the questionnaire data to be analyzed from the intelligent inspection system. Semantic reasoning is then performed on each question-answer pair in the questionnaire data to obtain the sentiment tendency of the question-answer pair. Based on the sentiment tendency, the emotional tendency expressed by the question-answer pair is determined as positive or negative. The analysis result for each question-answer pair is then determined, and the analysis result is used to determine whether there are safety hazards in the inspection items associated with the questions. Therefore, it is possible to quickly analyze questionnaire data of different formats and contents to determine whether there are safety hazards in the inspection items associated with the questions in the questionnaire data, improving the efficiency and accuracy of questionnaire data analysis. Attached Figure Description

[0016] Figure 1 This is a flowchart of an intelligent inspection option anomaly identification method according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the structure of an intelligent inspection option anomaly identification device according to Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the structure of an intelligent inspection option anomaly identification device according to Embodiment 3 of the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solution of the present invention, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0018] It is understood that the specific embodiments and accompanying drawings described herein are merely for explaining the invention and are not intended to limit the invention.

[0019] It is understood that, without conflict, the various embodiments and features in the embodiments of the present invention can be combined with each other.

[0020] It is understood that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, while the parts unrelated to the present invention are not shown in the drawings.

[0021] It is understood that each unit or module involved in the embodiments of the present invention may correspond to only one entity structure, or may be composed of multiple entity structures, or multiple units or modules may be integrated into one entity structure.

[0022] It is understood that the terms "first," "second," etc., in the embodiments of the present invention are used to distinguish different objects or to distinguish different treatments of the same object, rather than to describe a specific order of objects.

[0023] It is understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of this invention may occur in a different order than that marked in the accompanying drawings.

[0024] It is understood that the flowcharts and block diagrams of this invention illustrate the possible architecture, functions, and operations of systems, apparatuses, devices, and methods according to various embodiments of this invention. Each block in the flowchart or block diagram may represent a unit, module, program segment, or code, containing executable instructions for implementing the specified function. Furthermore, each block or combination of blocks in the block diagram and flowchart can be implemented using a hardware-based system to achieve the specified function, or using a combination of hardware and computer instructions.

[0025] It is understood that the units and modules involved in the embodiments of the present invention can be implemented by software or by hardware. For example, the units and modules can be located in a processor.

[0026] It should be noted that the scenario diagrams described in the embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided in the embodiments of this application. As those skilled in the art will know, with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0027] Data center server rooms are core IT production sites for storing, managing, and transmitting data, supporting enterprise business operations. Strict requirements govern many aspects of server rooms, including environment, hygiene, security, equipment emissions, and safety. For example, the temperature and humidity inside the server room must be strictly controlled within a specific range. The supply of mains power, backup generators, and related consumables such as fuel must be guaranteed. Rack placement, equipment cabling, and labeling must strictly adhere to standard configurations. Strict requirements also apply to the cleanliness of server bays, equipment maintenance, and fire safety equipment. Therefore, to ensure that all requirements are met, server room maintenance personnel or inspection robots conduct regular inspections and provide feedback on the inspection results through completed questionnaires. Server room management personnel then make corresponding maintenance decisions based on the inspection records, ensuring timely maintenance of the server room facilities as needed.

[0028] To achieve unified management of the operational status of data center inspections and to ensure efficient resource allocation and orderly supervision, it is necessary to collect and aggregate inspection questionnaire data from each data center and to uniformly analyze and control the operational status of each data center to ensure their safe operation. However, due to historical reasons, different data centers set different inspection indicators in the questionnaires according to their own needs and maintained them independently, resulting in a wide variety of aggregated questionnaires. This leads to time-consuming and costly data analysis.

[0029] Based on the problems existing in the current technology, this application proposes an intelligent inspection option anomaly identification method to improve the efficiency of questionnaire data analysis and processing.

[0030] Example 1: This embodiment provides a method for identifying abnormalities in intelligent inspection options, such as... Figure 1 As shown, the method includes the following steps S101 to S103: S101: Obtain the questionnaire data to be analyzed from the intelligent inspection system. The questionnaire data to be analyzed includes at least one question-answer pair consisting of a question and an answer.

[0031] Among them, the intelligent inspection system can collect inspection questionnaires from each computer room and extract the questionnaire data.

[0032] Specifically, the inspection questionnaire in the intelligent inspection system includes at least the question stem, question options, and user-filled answers, as shown in Table 1 below. Question-answer pairs can be extracted from the questionnaire data. A question and its corresponding user-filled answer constitute a question-answer pair.

[0033] Table 1: Sample Questions for Questionnaire 1

[0034]

[0035] Taking the sample inspection questionnaire data in Table 1 above as an example, the question "Fire extinguisher status" and the corresponding user's answer "normal" constitute a question-answer pair.

[0036] S102: Perform semantic reasoning on each question-and-answer pair to obtain the sentiment tendency of the question-and-answer pair.

[0037] Sentiment tendency is used to characterize whether the emotion expressed in the question and answer is positive or negative. Sentiment tendency reflects the user's viewpoint, attitude, or emotional tone conveyed regarding the question.

[0038] In this application, the sentiment of the question-and-answer pair can be determined based on the inspection of a specific item, and whether the status of the inspected item is positive or negative.

[0039] Specifically, semantic reasoning can determine the sentiment tendency of each question-and-answer pair. For example, consider the question-and-answer pair: Question: "Should the air conditioner be cleaned?", Answer: "Yes". Semantic reasoning on this question-and-answer pair reveals that in a data center scenario, cleanliness is crucial. Therefore, cleaning the air conditioner can be considered a positive action in a scenario where cleanliness is paramount. Thus, this question-and-answer pair has a positive sentiment tendency.

[0040] In this embodiment, the method for performing semantic reasoning on the question-answer pair to be tested can be Natural Language Processing (NLP), neural network models, machine learning, or other methods.

[0041] S103: Determine the analysis results for each question-and-answer pair based on the sentiment tendency of each pair.

[0042] The analysis results are used to characterize whether there are safety hazards in the inspection items associated with the problem.

[0043] Specifically, since the questions in the inspection questionnaire are usually designed based on whether there are security risks in the computer room, in order to record the operation of the computer room during the inspection process, the presence of security risks in the inspection items associated with the questions is determined manually or by template matching based on the questions and corresponding answers (i.e., question-answer pairs). However, the questionnaires for different computer rooms may be designed based on different inspection items and different security standards, resulting in significant differences in the question and option settings. This leads to inefficiency when the central computer room needs to analyze each question individually when summarizing and analyzing the inspection questionnaire data from various computer rooms. Therefore, in this embodiment, in order to quickly analyze different inspection questionnaire data, after extracting the question-answer pairs to be tested from the inspection questionnaire data, semantic reasoning can be performed on the question-answer pairs to determine the sentiment tendency of each question-answer pair. The analysis results of the inspection questionnaire data are determined based on the sentiment tendency of each question-answer pair. Based on the sentiment tendency of each question-answer pair, it can be determined whether there are security risks in the inspection items associated with the questions in the question-answer pairs. Thus... By performing semantic reasoning on each question-and-answer pair to determine its sentiment tendency, it is possible to identify any safety hazards associated with the inspection items related to each question-and-answer pair. This eliminates the need to analyze each question individually for questionnaires with different formats, thereby improving the efficiency of inspection questionnaire data analysis and reducing the cost of analysis.

[0044] In this embodiment, the first step is to acquire the inspection questionnaire data to be analyzed. Semantic reasoning is then performed on each question-and-answer pair in the questionnaire data to obtain the sentiment tendency of the question-and-answer pair. Based on the sentiment tendency, it is determined whether the sentiment expressed by the question-and-answer pair is positive or negative. The analysis results for each question-and-answer pair are then determined, and the results are used to determine whether there are any safety hazards in the inspection items associated with the questions. Therefore, it is possible to quickly analyze questionnaire data of different formats and contents, determine whether there are any safety hazards in the inspection items associated with the questions in the questionnaire data, and improve the efficiency and accuracy of questionnaire data analysis.

[0045] Optionally, S102 may specifically include the following steps: For each question-answer pair to be tested, the answer to the question-answer pair is matched with preset keywords to obtain the matching results; If the matching results indicate that the answer to the question-and-answer pair matches the preset keywords, the sentiment tendency of the question-and-answer pair is determined based on the preset keywords. If the matching results indicate that the answer to the question-and-answer pair does not match the preset keywords, the question-and-answer pair is input into a pre-trained semantic reasoning model to obtain the sentiment tendency of the question-and-answer pair.

[0046] The preset keywords are keywords set according to actual application scenarios that can directly determine the sentiment tendency of the question-and-answer pair. The sentiment tendency corresponding to each preset keyword is predetermined, and these can be words that appear frequently in questionnaires for specific scenarios (i.e., words used in answers to questionnaire questions with a unified strong sentiment, see the words in the question options in Table 2 below). In a specific scenario (such as a data center inspection scenario), if the preset keywords appear in the answers, the sentiment tendency of the question-and-answer pair can be directly determined as positive or negative. Examples include words such as "normal," "abnormal," "no abnormality," and "abnormality."

[0047] Specifically, in this embodiment, the answers to the question-and-answer pairs in the inspection questionnaire data are first matched with preset keywords. If they match the preset keywords, the sentiment tendency of the question-and-answer pair can be directly determined (examples of such question-and-answer pairs are shown in the questionnaire sample in Table 2 below). If they do not match the preset keywords, the question-and-answer pair is input into a pre-trained semantic reasoning model, and the sentiment tendency of the question-and-answer pair is quickly obtained through the semantic reasoning model. This allows for the rapid determination of the sentiment tendency of each question-and-answer pair, thereby improving the data analysis efficiency of the inspection questionnaire.

[0048] In one example, as shown in Table 2 of the questionnaire sample below, the answers to the question-answer pairs can be directly matched with preset keywords to determine the sentiment tendency of the question-answer pairs to be tested. Specifically, for the questions in the question-answer pairs corresponding to the question content, the user fills in the answers corresponding to the questions in the question-answer pairs to be tested.

[0049] Table 2. Sample Questionnaire Questions with a Strong Orientation Towards Uniformity in Options

[0050] In this embodiment, the semantic reasoning model can be trained based on historical questionnaire data.

[0051] In one example, suppose the preset keyword includes "abnormal" and the corresponding sentiment is negative. If the question-and-answer pair to be tested is: Question: "Fire extinguisher status", Answer: "abnormal", and the answer is matched with the preset keyword, then the answer matches the preset keyword "abnormal". Therefore, it can be directly determined that the sentiment of the question-and-answer pair to be tested is negative.

[0052] In practice, due to the different options in the question settings, some question-answer pairs cannot be directly determined by matching keywords in the answers to determine their sentiment. In this case, the question-answer pair can be input into a semantic reasoning model to predict the sentiment of the question-answer pair.

[0053] Optionally, the above process of matching the answers to the question-and-answer pairs with preset keywords to obtain matching results may include: The answers to the question-and-answer pairs are matched with preset keywords using regular expressions to obtain the matching results.

[0054] Regular expressions are tools used for matching, searching, and manipulating text. They use specific syntax and patterns to describe patterns in strings and can perform operations such as searching, replacing, and extracting from text.

[0055] Specifically, firstly, a regular expression pattern for a preset keyword can be defined (for example, metacharacters and character classes can be used to describe the pattern of the preset keyword, such as using "|" to represent an OR relationship and using square brackets [] to represent a character range); then, the answers to the question-answer pair to be tested are traversed, and the regular expression pattern of the preset keyword is applied to each answer for matching; if a match is successful, it means that the answer matches the preset keyword, and the matching result is output.

[0056] Optionally, the pre-trained semantic reasoning model mentioned above includes a first reasoning model and a second reasoning model. The process of inputting the question-and-answer pair into a pre-trained semantic reasoning model to obtain the sentiment tendency of the question-and-answer pair can include the following steps: Input the correct answer to the question into the first inference model to obtain the first inference result; If the first reasoning result includes the sentiment tendency of the answer, the sentiment tendency of the correct answer to the question to be tested is determined as the sentiment tendency of the correct answer to the question to be tested; If the first reasoning result does not include the sentiment of the answer, the question-answer pair to be tested is input into the second reasoning model to obtain the sentiment of the question in the question-answer pair and the logical constraint value of the answer. The logical constraint value of the answer is the preset value corresponding to the explicit definition of the answer in a logical sense, and the logical constraint value is 0 or 1. The sentiment of the question-and-answer pair is determined based on the sentiment of the questions and the logical constraints of the answers.

[0057] The logical constraint value is the preset value corresponding to the explicit definition of the answer in a logical sense. It can be preset; for example, if the logical meaning is affirmative, the preset value is 1, and if the logical meaning is negative, the preset value is 0.

[0058] Specifically, some question-answer pairs cannot be directly matched with preset keywords to determine sentiment bias. However, the sentiment bias of the question-answer pairs can be determined by reasoning from the answers (i.e., questionnaire questions with inconsistent options but containing bias), as shown in the questionnaire data example in Table 3 below. Therefore, for this type of question-answer pair, the answer can be directly input into the first reasoning model to obtain the sentiment bias of the answer (thus determining the sentiment bias of the question-answer pair).

[0059] Another type of question-and-answer pair exists where the options do not contain any emotional bias (e.g., words like "yes" and "no," which are only logically defined). Therefore, the emotional bias cannot be inferred solely from the answers. For this type of question-and-answer pair (i.e., questions with strong logical options requiring integration with the question stem), a second inference model is used to obtain the question's emotional bias and the logical constraints of the answers. Combining these two factors allows for the deduction of the question-and-answer pair's emotional bias. This is illustrated in the questionnaire data example shown in Table 4. Therefore, for question-and-answer pairs where the emotional bias cannot be inferred solely from the answers, the entire pair can be input into the second inference model to obtain its emotional bias.

[0060] In this embodiment, for question-and-answer pairs whose answers contain sentiment bias, only the answers of the question-and-answer pair need to be input into the first inference model to determine the sentiment bias; for question-and-answer pairs whose answers do not contain sentiment bias, the entire question-and-answer pair can be input into the second inference model to determine the sentiment bias. Therefore, different inference models can be used to determine sentiment bias based on different types of question-and-answer pairs, thereby improving the accuracy of sentiment bias determination.

[0061] In one example, the questionnaire questions with inconsistent and biased options are shown in Table 3 below.

[0062] Table 3: Sample Questionnaire Questions with Inconsistent Options and Prejudiced Bias

[0063] In one example, as shown in Table 4 below, there are problem examples where the options are logically sound and the result needs to be derived by combining the content of the question stem.

[0064] Table 4: Examples of questions where the options are logically sound and the answer requires combining the question stem with the relevant information.

[0065] Optionally, determining the sentiment tendency of the question-and-answer pair based on the sentiment tendency of the questions and the logical constraints of the answers may include: If the sentiment of the question is positive and the logical constraint value of the answer is 1, or if the sentiment of the question is negative and the logical constraint value of the answer is 0, the sentiment of the question to be tested is determined to be positive. If the sentiment of the question is negative and the logical constraint value of the answer is 1, or if the sentiment of the question is positive and the logical constraint value of the answer is 0, the sentiment of the question to be tested is determined to be negative.

[0066] Specifically, for question-and-answer pairs where the sentiment tendency cannot be directly determined by reasoning about the answers, the following approach is taken: first, determine the sentiment tendency corresponding to the question content in the question-and-answer pair; then, determine the logical constraint value corresponding to the answer; and finally, combine the sentiment tendency of the question and the logical constraint value of the answer to comprehensively determine the sentiment tendency of the question-and-answer pair. This allows for accurate determination of the sentiment tendency of the question-and-answer pair.

[0067] Optionally, the first inference model described above can be trained using the following steps: Obtain historical questionnaire data, which includes multiple historical question-and-answer pairs; The first question-and-answer pair was selected from multiple historical question-and-answer pairs, and the answers of the first question-and-answer pair have a sentimental bias. The sentiment of each correct answer in the first question is labeled to obtain the first feature information label; Construct the first training sample based on each first question-answer pair and its corresponding first feature information label; Based on multiple initial training samples, a pre-defined semantic reasoning model is trained until the training stopping condition is met, resulting in the trained initial reasoning model. The historical questionnaire data can be collected by the intelligent inspection system or by other means.

[0068] Pre-defined semantic reasoning models can be built using deep learning models, such as recurrent neural networks (RNNs) or convolutional neural networks (CNNs).

[0069] Specifically, firstly, the first question-answer pairs (i.e., historical question-answer pairs with sentimental answers) are obtained from historical questionnaire data; the sentimental orientation of the answers in each first question-answer pair is labeled to obtain the first feature information label; a first question-answer pair and its corresponding first feature information label are combined to form a training sample. All first question-answer pairs and their corresponding first feature information labels constitute a training sample set (i.e., the first training sample); the first training sample is input into a preset semantic reasoning model to obtain the first predicted feature information of the first question-answer pair; based on the first predicted feature information and the first feature information label, the loss function value of the semantic reasoning model is determined; if the loss function value does not meet the training stopping condition, the model parameters of the preset semantic reasoning model are adjusted, and the preset semantic reasoning model with adjusted parameters is trained using the first training sample set until the training stopping condition is met, thus obtaining the first reasoning model.

[0070] In this embodiment, first question-answer pairs (i.e., historical question-answer pairs with sentiment bias) can be obtained from historical questionnaire data, and the answers of the first question-answer pairs are labeled with sentiment bias. Then, first training samples are constructed based on each first question-answer pair and its corresponding sentiment bias label. Finally, a preset semantic reasoning model is trained using these training samples until the training stopping condition is met, thus arriving at the first reasoning model.

[0071] Optionally, the second inference model described above can be trained using the following steps: The second question-and-answer pair was selected from multiple historical question-and-answer pairs. The questions in the second question-and-answer pair have an emotional bias, and the answers have logical constraints. For each second question-and-answer pair, the sentiment tendency of the question and the logical constraint value of the answer are labeled to obtain the second feature information label; Construct second training samples based on each second question-answer pair and its corresponding second feature information label; Based on multiple second training samples, a pre-defined semantic reasoning model is trained until the training stopping condition is met, resulting in a trained second reasoning model.

[0072] Specifically, the training steps for the second inference model are the same as those for the first inference model, so they will not be repeated here.

[0073] In this embodiment, a first inference model and a second inference model can be trained using historical questionnaire data. These models can be used to predict and infer the sentiment of new question-answer pairs, thereby improving the efficiency and accuracy of the inspection questionnaire data processing.

[0074] Optionally, S103 above may specifically include the following steps: If the sentiment of the question-and-answer pair is positive, then the inspection items associated with the questions in the question-and-answer pair are determined to have no safety hazards. When the sentiment of the question-and-answer pair is negative, it is determined that the inspection items associated with the questions in the question-and-answer pair have safety hazards.

[0075] The questionnaire data processing method provided in this invention first acquires the questionnaire data to be analyzed from an intelligent inspection system. Semantic reasoning is then performed on each question-and-answer pair in the questionnaire data to obtain the sentiment tendency of the question-and-answer pair. Based on the sentiment tendency, it is determined whether the sentiment expressed by the question-and-answer pair is positive or negative. The analysis result for each question-and-answer pair is then determined, and the analysis results are used to determine whether there are any safety hazards in the inspection items associated with the questions. Therefore, it is possible to quickly analyze questionnaire data of different formats and contents, determine whether there are any safety hazards in the inspection items associated with the questions in the questionnaire data, and improve the efficiency and accuracy of questionnaire data analysis.

[0076] In one specific embodiment, the questionnaire data processing method may include the following steps: Step 1: Data Access and Data Exploration 1. Access historical questionnaire data from the business system via FTP file transfer protocol; 2. By reading historical questionnaire data using pandas and analyzing the distribution of question content (stem) and option settings in different questionnaires, it can be found that the questions in the questionnaires are mainly divided into three categories: (1) The conclusion can be drawn directly from the option settings, and the option settings are uniform, as shown in Table 2 above; (2) The conclusion can be drawn directly from the option settings, but the option settings are not uniform (equivalent to the first question and answer pair above), as shown in Table 3 above; (3) The conclusion cannot be drawn from the option settings, and it is necessary to analyze the options and question stem content together to draw a conclusion (equivalent to the second question and answer pair above). As shown in Table 4 above.

[0077] 3. Questionnaire Question Hazard Identification Task Analysis

[0078] (1) As shown in Table 2 above, the conclusion can be drawn directly based on the option settings. For example, if the question option is "normal", it can be considered that there is no hidden danger and the result is 1. If the question option is "abnormal", it can be considered that there is a hidden danger and the result is 1. Such questionnaire questions can directly draw results without model prediction.

[0079] (2) As shown in Table 3 above, it is necessary to judge whether there is a hidden danger based on the positive or negative content of the option (i.e. the emotional tendency of the answer). Since the options are varied, it is necessary to use an AI model to predict the tendency of the question options (using the first reasoning model for reasoning).

[0080] (3) As shown in Table 4 above, it is not possible to determine whether there is a hidden danger based solely on the content of the options. It is necessary to derive the logical value from the content of the options. The value is 1 when the option is "yes" or "there is", and 0 when it is "no" or "no". The value is set to 1 when the question content is positive and 0 when it is negative. When both the question content and the logical value of the options are 1 or both are 0, it is considered that there is no hidden danger. Otherwise, it is considered that there is a hidden danger. In this case, it is necessary to use an AI model to make a bias prediction on the question content (using the second reasoning model for reasoning).

[0081] Step 2: Data Processing and Feature Engineering (Obtaining Model Training Samples)

[0082] 1. For the questions and options in categories 2 and 3, perform text segmentation, stop word removal, and text cleaning. (This is equivalent to extracting historical question-and-answer pairs.)

[0083] 2. For the third type of question, it is necessary to extract the content from the question stem and analyze it to identify the tendency of the question, such as whether it is abnormal, whether there is a problem, whether it is normal, or whether there is no problem. If the first two options result in an abnormal outcome, output 0; if the latter two options result in a normal outcome, output 1. Table 5 below shows a sample table of content extraction for the question stem of the third type of question.

[0084] Table 5: Sample Table of Question Stem Content Extraction for Type 3 Questions

[0085] 3. Tag the options for Category 2 questions and the content extracted from Category 3 questions. Based on the question stem and options, and combined with business knowledge from DC and network specializations, assign labels. A positive sentiment is labeled as 1, and a negative sentiment as 0. (Hypersentiment labeling is performed on historical question-and-answer pairs to obtain sample data.)

[0086] 4. Data Augmentation: Examine the distribution of data with labels 1 and 0, and imput the sub-label data (upsampling or data augmentation), such as synonym replacement or swapping order. This can improve the prediction accuracy of the trained model.

[0087] 5. Split the historical questionnaire data into training and test sets.

[0088] Step 3: Model Training and Optimization

[0089] 1. The model is trained on the training set data using a deep neural network algorithm, and then the trained model is evaluated and its parameters are adjusted using the test set data.

[0090] 2. When the model evaluation accuracy is low, the model performance can be improved by adjusting parameters such as the learning rate, batch size, and number of training iterations.

[0091] Step 4: Model Engineering and Deployment

[0092] 1. Deploy the trained model through the cloud platform. Due to the large amount of data accessed by customers, increase concurrency by expanding the number of nodes (pods) to increase the training and inference speed of the accessed data. Here, the number of nodes is 20. 2. Deploy the engineered model on the cloud platform - China Unicom Network AI Platform as a cloud service.

[0093] Step 5: Real-time Questionnaire Question Model Reasoning (Using the trained reasoning model to perform data analysis on real-time questionnaire questions)

[0094] Specifically, the steps include the following: S201. Obtain real-time questionnaire data S202, Data Analysis Data analysis is performed on real-time questionnaire data to obtain the question-answer pairs to be tested.

[0095] S203, Analysis of Question Options

[0096] (1) Determine the correct answers to the questionnaire questions (i.e., the correct answers to the questions to be answered).

[0097] If the option content is matched using regular expressions and contains clear logical limitations and biases such as "no abnormality," "no abnormality," or "abnormal," and there is no semantic ambiguity, the inference result is directly output as 1 or 0. The questionnaire questions here are categorized as Type 1 questions.

[0098] (2) When it is impossible to match obvious logical constraints and tendencies through regular expressions, the AI ​​of the previously trained option content is used to make predictions (equivalent to reasoning the answer to the question and answer in the first reasoning model). Positive results are recorded as 1, negative results as 0, and the reasoning result is directly output as 1 or 0.

[0099] (3) When the question options are only logically limited, such as "yes," "no," "have," "none," or "not have," the reasoning result cannot be directly derived because the question content may be positive or negative when the option is logically 1. Therefore, it is necessary to combine the question content and the option content to derive the result together, and this type of question is identified as the third type of question. The question content is matched with the content after the question using regular expressions, and then the AI ​​predicts the tendency of the question content (equivalent to reasoning about the question-answer pair under test using the second reasoning model). When both the question content and the option content are positive or negative, the result is 1; when the question content and the option content have opposite tendencies, the result is 0.

[0100] Examples of the model inference results are shown in Tables 6 to 8 below. Table 6 corresponds to the first type of problem, Table 7 corresponds to the second type of problem, and Table 8 corresponds to the third type of problem.

[0101] Table 6: Examples of Reasoning Results for Type 1 Problem Models

[0102] Table 7: Examples of Reasoning Results for Type 2 Problem Models

[0103] Table 8: Examples of Reasoning Results for Type 3 Problem Models

[0104] In this embodiment, the numerous and varied questionnaire questions for data center inspection tasks can be divided into three categories. The question stems of the third category are extracted using regular expressions to identify potential issues. The options for the second category of questions and the extracted question stems from the third category are then labeled with the data center's and network experts' experience and logical tendencies. After training with historical questionnaire data, a neural network model that can predict the existence of potential hazards is obtained.

[0105] Example 2: like Figure 2 As shown, this embodiment provides an intelligent inspection option anomaly identification device for performing the above-described intelligent inspection option anomaly identification method, including: The first acquisition module 201 is used to acquire the questionnaire data to be analyzed from the intelligent inspection system. The questionnaire data to be analyzed includes at least one question-answer pair consisting of questions and answers. The reasoning module 202 is used to perform semantic reasoning on each question-answer pair to obtain the sentiment tendency of the question-answer pair. The sentiment tendency is used to characterize whether the sentiment expressed by the question-answer pair is positive or negative. The first determining module 203 is used to determine the analysis result of each question-and-answer pair based on the sentiment tendency of each question-and-answer pair. The analysis result is used to characterize whether there are safety hazards in the inspection items associated with the question.

[0106] Optionally, the inference module 202 mentioned above includes: The matching unit is used to match the answer of each question-answer pair with preset keywords to obtain the matching result; The first determining unit is used to determine the sentiment tendency of the question-and-answer pair based on the preset keywords when the matching result indicates that the answer of the question-and-answer pair matches the preset keywords. The reasoning unit is used to input the question-and-answer pair into a pre-trained semantic reasoning model when the matching result indicates that the answer of the question-and-answer pair does not match the preset keywords, so as to obtain the sentiment tendency of the question-and-answer pair.

[0107] Optionally, the above-mentioned reasoning unit includes: The first reasoning subunit is used to input the answer of the question-and-answer pair to the first reasoning model to obtain the first reasoning result; The first determining subunit is used to determine the sentiment tendency of the answer of the question-and-answer pair as the sentiment tendency of the question-and-answer pair when the first reasoning result includes the sentiment tendency of the answer. The second reasoning subunit is used to input the question-and-answer pair to the second reasoning model when the first reasoning result does not include the sentiment of the answer, to obtain the sentiment of the question and the logical constraint value of the answer in the question-and-answer pair. The logical constraint value of the answer is a preset value corresponding to the explicit definition of the answer in a logical sense. The logical constraint value is 0 or 1. The second determining subunit is used to determine the sentiment tendency of the question-answer pair based on the sentiment tendency of the question and the logical constraint value of the answer in the question-answer pair to be tested.

[0108] In this embodiment, the questionnaire data to be analyzed from the intelligent inspection system is first acquired. Semantic reasoning is then performed on each question-and-answer pair in the questionnaire data to obtain the sentiment tendency of the question-and-answer pair. Based on the sentiment tendency, it is determined whether the sentiment expressed by the question-and-answer pair is positive or negative. The analysis result of each question-and-answer pair is then determined, and the analysis result is used to determine whether there are any safety hazards in the inspection items associated with the questions. Therefore, questionnaire data of different formats and contents can be quickly analyzed to determine whether there are any safety hazards in the inspection items associated with the questions in the questionnaire data, improving the efficiency and accuracy of questionnaire data analysis.

[0109] Example 3: Figure 3 A schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application is shown.

[0110] An electronic device may include a processor 301 and a memory 302 storing computer program instructions.

[0111] Specifically, the processor 301 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0112] Memory 302 may include mass storage for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 302 may include removable or non-removable (or fixed) media. Where appropriate, memory 302 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 302 is non-volatile solid-state memory.

[0113] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the methods according to one aspect of this disclosure.

[0114] The processor 301 reads and executes computer program instructions stored in the memory 302 to implement any of the intelligent inspection option anomaly identification methods in the above embodiments.

[0115] In one example, the electronic device may also include a communication interface 303 and a bus 304. For example, Figure 3 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 304 and complete communication with each other.

[0116] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0117] Bus 304 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 304 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0118] Furthermore, in conjunction with the intelligent inspection option anomaly identification method in the above embodiments, this application embodiment can provide a computer storage medium for implementation. This computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the barcode recognition methods in the above embodiments.

[0119] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0120] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0121] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0122] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in 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, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0123] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for identifying anomalies in intelligent inspection options, characterized in that, include: Acquire the questionnaire data to be analyzed from the intelligent inspection system, wherein the questionnaire data to be analyzed includes at least one question-answer pair consisting of a question and an answer; Semantic reasoning is performed on each of the question-and-answer pairs to obtain the sentiment tendency of the question-and-answer pairs to be tested. The sentiment tendency is used to characterize whether the sentiment expressed by the question-and-answer pairs to be tested is positive or negative. Based on the sentiment tendency of each question-and-answer pair, the analysis result of each question-and-answer pair is determined. The analysis result is used to characterize whether there are safety hazards in the inspection items associated with the question. The step of performing semantic reasoning on each of the question-and-answer pairs to obtain the sentiment tendency of the question-and-answer pairs includes: For each question-answer pair to be tested, the answer of the question-answer pair is matched with preset keywords to obtain the matching result; If the matching result indicates that the answer of the question-and-answer pair matches the preset keywords, the sentiment tendency of the question-and-answer pair is determined based on the preset keywords. If the matching result indicates that the answer of the question-and-answer pair does not match the preset keywords, the question-and-answer pair is input into a pre-trained semantic reasoning model to obtain the sentiment tendency of the question-and-answer pair. The pre-trained semantic reasoning model includes a first reasoning model and a second reasoning model. The step of inputting the question-and-answer pair to be tested into a pre-trained semantic reasoning model to obtain the sentiment tendency of the question-and-answer pair includes: The answers to the question-and-answer pairs to be tested are input into the first reasoning model to obtain the first reasoning result; If the first reasoning result includes the sentiment tendency of the answer, the sentiment tendency of the answer of the question-and-answer pair to be tested is determined as the sentiment tendency of the question-and-answer pair to be tested. If the first reasoning result does not include the sentiment of the answer, the question-answer pair to be tested is input into the second reasoning model to obtain the sentiment of the question and the logical constraint value of the answer in the question-answer pair to be tested. The logical constraint value of the answer is the preset value corresponding to the explicit definition of the answer in a logical sense. The logical constraint value is 0 or 1. The sentiment of the question-and-answer pair is determined based on the sentiment tendency of the questions and the logical constraints of the answers in the question-and-answer pair to be tested.

2. The method according to claim 1, characterized in that, The step of determining the sentiment tendency of the question-and-answer pair based on the sentiment tendency of the questions and the logical constraint values ​​of the answers includes: If the sentiment of the question is positive and the logical constraint value of the answer is 1, or if the sentiment of the question is negative and the logical constraint value of the answer is 0, then the sentiment of the question to be tested is determined to be positive. If the sentiment of the question is negative and the logical constraint value of the answer is 1, or if the sentiment of the question is positive and the logical constraint value of the answer is 0, then the sentiment of the question to be tested is determined to be negative.

3. The method according to claim 1, characterized in that, Before inputting the question-answer pair into a pre-trained semantic reasoning model to obtain the sentiment tendency of the answers to the question-answer pair, the method further includes: Obtain historical questionnaire data, which includes multiple historical question-answer pairs; The first question-and-answer pair is selected from the multiple historical question-and-answer pairs, and the answers of the first question-and-answer pair have a sentiment bias. The sentiment of each correct answer in the first question is labeled to obtain the first feature information label; Construct a first training sample based on each of the first question-answer pairs and its corresponding first feature information label; Based on multiple first training samples, a preset semantic reasoning model is trained until the training stopping condition is met, and the trained first reasoning model is obtained. A second question-and-answer pair is selected from the multiple historical question-and-answer pairs. The questions in the second question-and-answer pair have an emotional bias, and the answers have logically limiting significance. For each second question-and-answer pair, the sentiment tendency of the question and the logical constraint value of the answer are labeled to obtain the second feature information label; Construct a second training sample based on each second question-answer pair and its corresponding second feature information label; Based on multiple second training samples, a preset semantic reasoning model is trained until the training stopping condition is met, resulting in a trained second reasoning model.

4. The method according to claim 1, characterized in that, The step of matching the answers of the question-and-answer pair to be tested with preset keywords to obtain the matching results includes: The answers to the question-and-answer pairs to be tested are matched with preset keywords using regular expressions to obtain the matching results.

5. The method according to claim 1, characterized in that, The step of determining the analysis result for each question-and-answer pair based on its sentiment tendency includes: If the sentiment of the question-and-answer pair to be tested is positive, it is determined that there are no safety hazards in the inspection items associated with the questions in the question-and-answer pair to be tested. If the sentiment of the question-and-answer pair to be tested is negative, it is determined that the inspection item associated with the question in the question-and-answer pair has a safety hazard.

6. An intelligent inspection option anomaly identification device, characterized in that, include: The first acquisition module is used to acquire the questionnaire data to be analyzed from the intelligent inspection system. The questionnaire data to be analyzed includes at least one question-answer pair consisting of questions and answers. The reasoning module is used to perform semantic reasoning on each of the question-and-answer pairs to obtain the sentiment tendency of the question-and-answer pairs to be tested. The sentiment tendency is used to characterize whether the sentiment expressed by the question-and-answer pairs to be tested is positive or negative. The first determining module is used to determine the analysis result of each question-and-answer pair based on the sentiment tendency of each question-and-answer pair to be tested. The analysis result is used to characterize whether there are safety hazards in the inspection items associated with the question. The reasoning module includes: The matching unit is used to match the answer of each question-answer pair with preset keywords to obtain the matching result. The first determining unit is configured to determine the sentiment tendency of the question-and-answer pair based on the preset keywords when the matching result indicates that the answer of the question-and-answer pair matches the preset keywords. The reasoning unit is used to input the question-and-answer pair to a pre-trained semantic reasoning model to obtain the sentiment tendency of the question-and-answer pair when the matching result indicates that the answer of the question-and-answer pair does not match the preset keywords. The pre-trained semantic reasoning model includes a first reasoning model and a second reasoning model. The reasoning unit includes: The first reasoning subunit is used to input the answer of the question-and-answer pair to the first reasoning model to obtain the first reasoning result; The first determining subunit is used to determine the sentiment tendency of the answer of the question-and-answer pair as the sentiment tendency of the question-and-answer pair when the first reasoning result includes the sentiment tendency of the answer. The second reasoning subunit is used to input the question-and-answer pair to the second reasoning model when the first reasoning result does not include the sentiment of the answer, to obtain the sentiment of the question and the logical constraint value of the answer in the question-and-answer pair. The logical constraint value of the answer is a preset value corresponding to the explicit definition of the answer in a logical sense. The logical constraint value is 0 or 1. The second determining subunit is used to determine the sentiment tendency of the question-answer pair based on the sentiment tendency of the question and the logical constraint value of the answer in the question-answer pair to be tested.

7. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the intelligent inspection option anomaly identification method as described in any one of claims 1-5.

Images