Automatic satisfaction data processing method and system based on big data resources

Abstract

The invention relates to the field of text research data processing methods, in particular to a satisfaction data automatic processing method and system based on big data resources, and the method comprises the steps: obtaining satisfaction data; the data type of the satisfaction data is recognized, the data type comprises a digital type and a text type, and the satisfaction data is cleaned and analyzed according to the data type; modeling is carried out according to data types, data processing is carried out according to data of digital types through a general model, and sentiment analysis and sentiment mining are carried out according to data of text types; automatically analyzing relevance and trend in the satisfaction data, and outputting predicted user behaviors or satisfaction changes. According to the method, analysis errors caused by deviation of data are reduced, and an investigation result obtained through analysis can be more accurate.

Description

Technical Field

[0001] The present invention relates to the field of text survey data processing methods, and in particular to a method and system for automatically processing satisfaction data based on big data resources. Background Art

[0002] Research is the systematic and purposeful process of gathering information on usage, user experiences, opinions, and suggestions regarding various market activities, products, and services through various channels and methods. While research can focus on a variety of topics, such as satisfaction, user experience, new situations, and new markets, the most common and important research topic is satisfaction research.

[0003] Satisfaction surveys are mainly implemented through paper or electronic questionnaires. Questionnaires with set questions are distributed to the target location to obtain the survey data in the questionnaire. Corresponding survey results can be obtained by analyzing and processing the survey data. The analysis and processing of survey data is currently mainly carried out through behavioral graphs. For example, in the literature survey data processing methods, devices, and storage media, the following methods are included: selecting surveyed users related to the surveyed object, determining the surveyed user's predicted evaluation information on each surveyed object's survey attributes based on the surveyed user's user behavior graph, and determining the survey topic that is suitable for the surveyed user from the surveyed object's survey topic set based on the surveyed user's predicted evaluation information on each surveyed object's survey attributes, so as to form a personalized survey questionnaire suitable for the surveyed user. This improves the accuracy of the survey results and enhances the survey effect.

[0004] However, due to the inaccuracy of the survey data source, the data processing results are biased, making the survey results unreliable. Summary of the Invention

[0005] One of the purposes of the present invention is to provide a method for automatically processing satisfaction data based on big data resources to solve the problem of unreliable survey results.

[0006] The automatic processing method of satisfaction data based on big data resources in this solution includes:

[0007] Step 1: Obtain satisfaction data, which includes small sample survey data, big data crawled data, and synthetic data;

[0008] Step 2: Identify the data type of the satisfaction data, which includes numeric and text types, and clean and analyze the satisfaction data according to the data type;

[0009] Step 3: Model the data type. For numerical data, use a general model to process the data. For text data, use sentiment analysis and emotion mining.

[0010] Step 4: Automatically analyze the correlations and trends in the satisfaction data and output predicted user behavior or satisfaction changes.

[0011] The beneficial effects of this program are:

[0012] By combining a variety of survey data, including small sample data, big data, and synthetic data, and then cleaning and analyzing satisfaction data, different data types are processed differently. The data's correlations and trends are automatically analyzed to reveal changes in user behavior or satisfaction. This reduces analytical errors caused by inherent data bias, resulting in more accurate survey results.

[0013] Furthermore, in step 2, when the data type is a numeric type, data cleaning is performed through data unification, missing value supplementation and logical judgment. The data unification is to convert the multi-level evaluation standards into a percentage score, the missing value supplementation is to fill in the mean of the indicator satisfaction evaluation, and the logical judgment is to judge unreliable answer data based on the total coordination of the answer or the answer time being less than a threshold.

[0014] The beneficial effect is: different cleaning methods are performed according to the data type, so that the data can be processed in an automatic and unified process with reduced errors.

[0015] Furthermore, in step 2, when the data type is text, cleaning is performed according to the following steps:

[0016] Step 2.1: Label the text data into three categories: normal data, abnormal emotions, and random answers;

[0017] In step 2.2, we use the fine-tuned Reberta model to convert the text into embedding vectors. Then, we use the KMeans algorithm with cosine distance as the distance metric and the three label-related data to build a three-category clustering algorithm model.

[0018] In step 2.3, when new data is input into the system, the constructed three-category clustering model is used for classification prediction. The prediction results are fed back to the backend, and small-scale label corrections are made for samples with doubts or inaccurate classifications. The corrected labeled data is immediately used to train the three-category clustering model. By updating the cluster centers of the three-category clustering model online, that is, updating the three-category algorithm model, the classification accuracy and model adaptability are improved.

[0019] In step 2.4, the label obtained by the three-class clustering model is used as the final label.

[0020] The beneficial effect is that, for text-based data, data cleaning can make it more suitable for needs. Clustering algorithms can also be updated online. For example, after successfully building a three-category algorithm, when new data is uploaded, the original algorithm model can be used to predict the classification. The classified data is then sent to the backend and returned to the user. The user can then make small label corrections to samples with doubts or inaccurate classifications, and then use the corrected data to update the cluster centers of the built clustering algorithm model, thus updating the clustering model.

[0021] Furthermore, in step 2.2, for each piece of data, the text is converted into a vector using a fine-tuned Reberta model, and then the center point distance of the clustering algorithm in step 2.3 is calculated, and 1-center point distance is used as the similarity. At the same time, the obtained text embedding vector data is stored in the Es database.

[0022] The beneficial effect is that the data feature calculation can accurately extract the content in the text data to accurately and automatically process the survey data.

[0023] Furthermore, it also includes step 5, automatically optimizing the survey indicators in the satisfaction data in step 1 according to the output results in step 4, obtaining the most popular data based on big data surveys, and automatically optimizing in combination with survey requirements.

[0024] The beneficial effect is that by automatically optimizing the survey indicators, the survey data can be quickly adjusted in a timely manner according to the actual survey needs, thereby reducing the workload of updating and optimizing the survey data.

[0025] Furthermore, in step 5, by comparing data by year, data between different regions, and data at different levels, the data with the highest attention is retained as the target data for automatic optimization.

[0026] The beneficial effect is: optimizing the target data according to different comparison results, making the optimization results more accurate and more in line with actual research needs.

[0027] Furthermore, the method further includes step 6, in which, based on step 5, the survey indicators of different industries are automatically generated into the survey indicators of the target industry according to the industry relevance and the relevance of the satisfaction survey questions.

[0028] The beneficial effect is that the automatic generation of research indicators between different industries can simplify the workload of research work and make the research work more efficient.

[0029] Furthermore, in step 6, when the industry relevance is between 50% and 60%, if the relevance of the satisfaction survey question is greater than 90%, then the satisfaction survey question is used as the satisfaction survey question for the target industry; if the relevance of the satisfaction survey question is greater than 70% and less than 90%, then the satisfaction survey question is used as a quasi-satisfaction survey question for the target industry; if the relevance of the satisfaction survey question is less than 70%, then the satisfaction survey question is used as an alternative satisfaction survey question for the target industry;

[0030] When the industry relevance is between 60% and 80%, if the relevance of the satisfaction survey question is greater than 80%, then the satisfaction survey question will be used as the satisfaction survey question for the target industry; if the relevance of the satisfaction survey question is greater than 60% but less than 80%, then the satisfaction survey question will be used as the quasi-satisfaction survey question for the target industry; if the relevance of the satisfaction survey question is less than 60%, then the satisfaction survey question will be used as the alternative satisfaction survey question for the target industry;

[0031] When the industry relevance is between 80% and 100%, if the relevance of the satisfaction survey question is greater than 70%, then the satisfaction survey question will be used as the satisfaction survey question for the target industry; if the relevance of the satisfaction survey question is greater than 60% and less than 70%, then the satisfaction survey question will be used as the quasi-satisfaction survey question for the target industry; if the relevance of the satisfaction survey question is less than 60%, then the satisfaction survey question will be used as the alternative satisfaction survey question for the target industry.

[0032] The beneficial effect is: according to the relevance of different industries, different survey questions are determined based on the relevance of satisfaction survey questions, which improves the accuracy of survey indicator setting and is more in line with actual needs.

[0033] A second object of the present invention is to provide an automatic processing system for satisfaction data based on big data resources to improve the efficiency and accuracy of automatic processing of survey results.

[0034] The automatic processing system of satisfaction data based on big data resources includes a processor storing a computer executable program, and the processor processes the satisfaction data according to the automatic processing method of satisfaction data based on big data resources as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a flowchart of an embodiment of a method for automatically processing satisfaction data based on big data resources according to the present invention;

[0036] Figure 2 It is a schematic diagram of the Kano model in the automatic processing method of satisfaction data based on big data resources of the present invention. DETAILED DESCRIPTION

[0037] The following is further explained in detail through specific implementation methods.

[0038] Example 1

[0039] Automatic processing methods of satisfaction data based on big data resources, such as Figure 1 As shown, the following steps are included:

[0040] Step 1: Obtain survey satisfaction data. Survey data can be obtained through questionnaires. Different questionnaire questions can be set based on different industries, yielding different satisfaction data. For example, in patient satisfaction surveys, satisfaction scores for medical guides, registration, nurse service attitude, ward management, and other aspects can be collected. In transportation management, information such as traffic violations, interviewees, and conductor service attitude can be collected and recorded. Furthermore, information such as telecommunications company service satisfaction scores, the clarity of marketing plans, and expectations for telecommunications services can be collected. Satisfaction data includes small-sample survey data, big data crawled data, and synthetic data. Small-sample survey data is obtained through traditional survey methods, such as intercepted interviews, telephone interviews, and online surveys. The sample size is typically several hundred to several thousand. While the sample size is small, it is highly reliable. Big data crawled data uses crawler technology to collect online satisfaction data (consumer reviews, primarily text) by industry category to enrich the satisfaction analysis text library.

[0041] Synthetic data is created using computational methods and simulations. The results mimic the statistical properties of real-world data but do not contain actual real-world observations. The generated data can take various forms, including text, numbers, tables, or more complex types such as images and videos. Synthetic data techniques can be used to generate large amounts of satisfaction survey data based on a small amount of existing satisfaction survey data, optimizing the data structure and increasing the richness and diversity of the data.

[0042] Step 2: Identify the data type of the satisfaction data, which includes numeric type and text type, and clean and analyze the satisfaction data according to the data type.

[0043] When the data type is numeric, data cleaning is performed through data unification, missing value filling, and logical reasoning. Data unification involves converting multi-level evaluation criteria into a percentage score system. For example, satisfaction surveys often use a five- or ten-point rating system, with the five-point rating system (from high to low) consisting of very satisfied, relatively satisfied, average, dissatisfied, and very dissatisfied. The ten-point rating system consists of scores of 1 / 2 / 3 / 4 / 5 / 6 / 7 / 8 / 9 / 10. For ease of processing and understanding, each rating system must be converted to a percentage score system. For example, the five-point rating system consists of very satisfied (100 points), relatively satisfied (80 points), average (60 points), dissatisfied (40 points), and very dissatisfied (20 points). For the ten-point rating system, the corresponding score is multiplied by 10. Missing values are filled using the mean of the satisfaction ratings for each indicator. Missing values indicate that consumers are not familiar with a particular indicator in the survey and are unable to provide a satisfaction level response. Logical judgment is to identify unreliable answer data based on the overall coordination of answers or the response time being less than a threshold. If a consumer gives full marks or low marks throughout the survey, the consumer's satisfaction rating is considered questionable and will not be included in the calculation. If the consumer's response time is too long or too short, the rating will also be considered questionable.

[0044] When the data type is text, clean it according to the following steps:

[0045] In step 2.1, the text data is divided into three categories: normal data, abnormal emotions, and random answers. Labels are formed and classified using a pre-trained clustering model. The core is to calculate the distance from the survey data to each defined cluster center and assign it to the cluster with the shortest distance. This is used to determine whether the data is normal data, abnormal emotions, or random answers. The specific classification steps are as follows:

[0046] 2.1.1: For the initial data collection, use the clustering model to convert the text into an embedding vector;

[0047] 2.1.2: Use the kmeans clustering algorithm with cosine distance on the vectors in 2.1.1 to select the category that best matches the three labels, namely normal data, abnormal emotions, and random answers;

[0048] 2.1.3: Perform another clustering algorithm on the data selected in 2.1.2 to update the cluster centers of the constructed three-classification algorithm;

[0049] 2.1.4: Prediction: Use the above clustering model to predict the newly generated survey data. Compare the cosine distance between the above category center and the survey data, and classify it into the closest category.

[0050] While classifying new survey data, record all survey data with a similarity greater than 98%. As survey data with a similarity greater than 90% accumulates, regularly update the clustering model and coding model based on sufficient data volume to maintain the continued effectiveness of the model. Update the clustering model and coding model according to the following procedures:

[0051] A. First, use the fine-tuned Reberta model to convert the satisfaction survey data into a vector, for example, (3.1123, 5.22145, ..., 2.12442);

[0052] B. Then use the kmeans clustering algorithm with cosine distance to perform preliminary clustering on the preset proportion of satisfaction survey data and obtain the center point of each cluster;

[0053] C. Calculate the distances between the remaining satisfaction survey data (a preset proportion) and the center points of each cluster, and assign them to the point closest to them, which is 70% of the satisfaction survey data.

[0054] The preset ratio is 30%. Setting the preset ratio can reduce computing pressure and machine resource requirements.

[0055] D. In each cluster, select the data with the highest similarity of >95% as the positive sample, and the one with the lowest similarity as the negative sample;

[0056] E. Update negative samples: Randomly extract a piece of data from other clusters to calculate the similarity, and compare it with the similarity of the negative sample, and select the one with the lowest similarity as the negative sample.

[0057] In step 2.2, we use the fine-tuned Reberta model to calculate similarity, convert the text into vectors, and then use the KMeans algorithm with cosine distance as the distance metric and the three label-related data to build a three-category clustering model;

[0058] Specifically: a coding model based on ReBERT fine-tuning is used to extract the data features of the satisfaction data, and the cosine similarity algorithm is used to calculate the similarity between the data features of the data and the data features of the satisfaction data. The cosine similarity algorithm calculates the similarity between the label and the text; clustering algorithm: the kmeans algorithm with cosine distance is used, and the data is processed in batches. Pre-encoding: that is, encoding is performed when the text enters the database, and the corresponding text vector can be directly obtained from the es database during calculation.

[0059] In step 2.3, when new data is input into the system, the constructed three-category clustering model is used for classification prediction. The prediction results are fed back to the backend, and small-scale label corrections are made for samples with doubts or inaccurate classifications. The corrected labeled data is immediately used to train the three-category clustering model. By updating the cluster centers of the three-category clustering model online, that is, updating the three-category algorithm model, the classification accuracy and model adaptability are improved.

[0060] In step 2.4, the label obtained by the three-class clustering model is used as the final label.

[0061] Step 3: Model the data type and process the digital data with a general model. The general model is as follows: Figure 2 The existing Kano model shown above uses sentiment analysis and emotion mining for text-based data to control and optimize data quality. Data quality control and optimization consists of two main stages: during the survey and after the survey is completed. Specifically,

[0062] To control and optimize data quality during the survey process, monitor emotional instability and random responses: During survey data collection, pay special attention to the frequency of these responses. If the frequency of these responses exceeds a set threshold during a survey, take immediate action to optimize the survey process. This may involve adjusting questionnaire design, improving interviewer guidance, and improving communication skills.

[0063] In order to control and optimize the data quality after the survey is completed, data review and analysis are carried out. After each survey is completed, all collected data are reviewed, including normal data, abnormal emotional data and random answer data; the characteristics and causes of each category of data are analyzed to extract valuable information; methods or experiences that effectively reduce abnormal emotional and random answer data are summarized and applied to future survey practices, constantly exploring improvement paths to improve the overall quality of the survey.

[0064] Step 4: Automatically analyze the correlations and trends in the satisfaction data and output predicted user behavior or satisfaction changes.

[0065] In step 5, based on the output from step 4, the survey indicators in the satisfaction data from step 1 are automatically optimized. Based on the big data survey, the most popular data is obtained and automatically optimized according to the survey requirements. By comparing data by year, by region, and by level, the most popular data is retained as the target data for automatic optimization.

[0066] Step 6: Based on step 5, survey indicators for different industries are automatically generated according to the industry relevance and the relevance of satisfaction survey questions to generate survey indicators for the target industry.

[0067] Industries are classified by primary indicators and secondary indicators. The judgment of industry relevance is carried out according to the primary indicators first and then the secondary indicators. The primary indicators of the industry include sports, communications, real estate, automobiles, tobacco, transportation, urban management, finance, energy, the Disabled Persons' Federation, the Municipal Supervision Bureau, the National Health Commission, the Consumers Association, housing and construction, the cultural and tourism industry, etc. The secondary indicators include industries, venues, sports consumption, sports tourism statistics under sports, satisfaction, channel undercover visits, process penetration tests, complaints, consumer research, etc. under communications, mystery customers, satisfaction, customer research, brand research, business positioning, etc. under real estate, mystery customers, product research, customer research, etc. under automobiles, satisfaction, market monitoring, purification rate, etc. under tobacco, satisfaction, mystery customers, service capabilities, etc. under transportation, and daily assessment, garbage classification, urban management satisfaction, third-party review, etc. under urban management.

[0068] When judging the correlation, if there is a dependency between automobiles and transportation, the correlation judged based on the first-level indicators is 45%-50%, and the correlation is increased by 5% based on the overlap of the second-level indicators, then the correlation between the automobile industry and the transportation industry is 50%-55%.

[0069] When the industry relevance is between 50% and 60%, if the relevance of the satisfaction survey question is greater than 90%, then the satisfaction survey question will be used as the satisfaction survey question for the target industry; if the relevance of the satisfaction survey question is greater than 70% but less than 90%, then the satisfaction survey question will be used as the quasi-satisfaction survey question for the target industry; if the relevance of the satisfaction survey question is less than 70%, then the satisfaction survey question will be used as the alternative satisfaction survey question for the target industry;

[0070] When the industry relevance is between 60% and 80%, if the relevance of the satisfaction survey question is greater than 80%, then the satisfaction survey question will be used as the satisfaction survey question for the target industry; if the relevance of the satisfaction survey question is greater than 60% but less than 80%, then the satisfaction survey question will be used as the quasi-satisfaction survey question for the target industry; if the relevance of the satisfaction survey question is less than 60%, then the satisfaction survey question will be used as the alternative satisfaction survey question for the target industry;

[0071] When the industry relevance is between 80% and 100%, if the relevance of the satisfaction survey question is greater than 70%, then the satisfaction survey question will be used as the satisfaction survey question for the target industry; if the relevance of the satisfaction survey question is greater than 60% and less than 70%, then the satisfaction survey question will be used as the quasi-satisfaction survey question for the target industry; if the relevance of the satisfaction survey question is less than 60%, then the satisfaction survey question will be used as the alternative satisfaction survey question for the target industry.

[0072] Compared with the prior art, this embodiment combines a variety of different data such as small sample data, big data, and synthetic data, and then cleans and analyzes the data, unifies the data format and expression, facilitates subsequent processing, and reduces the deviation results caused by the inconsistency of the data itself. Different processing is performed for different data types, and the correlation and trend of the data are automatically analyzed to obtain changes in user behavior or satisfaction. Reducing the analysis error caused by the deviation of the data itself can make the survey results obtained by analysis more accurate. And on the basis of using the clustering algorithm to classify the survey data, using the method process and optimization scheme of this embodiment, according to the set goals, the ideal data should represent the normal and rational answers of the survey subjects. Therefore, the survey data is divided into three categories: normal data, abnormal emotions, and random answers. The main purpose is to exclude the latter two types of data to obtain high-quality survey results.

[0073] Example 2

[0074] The automatic processing system of satisfaction data based on big data resources is based on the method of embodiment one, and includes a processor storing a computer executable program. The processor processes the satisfaction data according to the automatic processing method of satisfaction data based on big data resources as described in embodiment one.

[0075] The above is only an embodiment of the present invention, and the common knowledge such as the specific structure and characteristics of the scheme is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. The automatic processing method of satisfaction data based on big data resources is characterized by: include: Step 1: Obtain satisfaction data, which includes small sample survey data, big data crawled data, and synthetic data; Step 2: Identify the data type of the satisfaction data, which includes numeric and text types, and clean and analyze the satisfaction data according to the data type; Step 3: Model the data type. For numerical data, use a general model to process the data. For text data, use sentiment analysis and emotion mining. Step 4: Automatically analyze the correlations and trends in the satisfaction data and output predicted user behavior or satisfaction changes.

2. The method for automatically processing satisfaction data based on big data resources according to claim 1 is characterized by: In step 2, when the data type is a numeric type, data cleaning is performed through data unification, missing value supplementation and logical judgment. The data unification is to convert the multi-level evaluation standards into a percentage score, the missing value supplementation is to fill in the mean of the indicator satisfaction evaluation, and the logical judgment is to judge unreliable answer data based on the total coordination of the answer or the answer time being less than a threshold.

3. The method for automatically processing satisfaction data based on big data resources according to claim 2, characterized in that: In step 2, if the data type is text, clean it according to the following steps: In step 2.1, we assign labels to the text data: normal data, abnormal emotions, and random answers. In step 2.2, we use a fine-tuned Reberta model to calculate similarity, convert the text into embedding vectors, and then use the KMeans algorithm with cosine distance as the distance metric to construct a clustering algorithm with three clusters, i.e., a three-class clustering model. In step 2.3, when new data is input into the system, the constructed three-category clustering model is used for classification prediction. The prediction results are fed back to the backend, and small-scale label corrections are performed on samples with doubts or inaccurate classifications. The corrected labeled data is immediately used to train the three-category clustering model. By updating the cluster center points of the three-category clustering model online, the three-category clustering model is updated. In step 2.4, the label obtained by the three-class clustering model is used as the final label.

4. The method for automatically processing satisfaction data based on big data resources according to claim 3 is characterized by: In step 2.2, for each piece of data, the text is converted into an embedding vector using a fine-tuned Reberta model, and then the center point distance with the clustering algorithm in step 2.3 is calculated, and 1-center point distance is used as the similarity, and the converted text embedding vector data is stored in the Es database at the same time.

5. The method for automatically processing satisfaction data based on big data resources according to claim 1 is characterized by: The method also includes step 5, which automatically optimizes the survey indicators in the satisfaction data in step 1 according to the output results in step 4, obtains the most popular data based on the big data survey, and automatically optimizes it in combination with the survey requirements.

6. The method for automatically processing satisfaction data based on big data resources according to claim 5 is characterized by: In step 5, by comparing data by year, data between different regions, and data at different levels, the data with the highest attention is retained as the target data for automatic optimization.

7. The method for automatically processing satisfaction data based on big data resources according to claim 6, characterized in that: It also includes step 6, which automatically generates the research indicators of the target industry based on the research indicators of different industries according to the industry relevance and the relevance of satisfaction survey questions on the basis of step 5.

8. The method for automatically processing satisfaction data based on big data resources according to claim 5 is characterized by: In step 6, when the industry relevance is between 50% and 60%, if the relevance of the satisfaction survey question is greater than 90%, then the satisfaction survey question is used as the satisfaction survey question for the target industry; If the relevance of a satisfaction survey question is greater than 70% and less than 90%, then the satisfaction survey question will be used as a quasi-satisfaction survey question for the target industry; If the relevance of a satisfaction survey question is less than 70%, then the satisfaction survey question will be used as an alternative satisfaction survey question for the target industry; When the industry relevance is between 60% and 80%, if the relevance of a satisfaction survey question is greater than 80%, then the satisfaction survey question will be used as the satisfaction survey question for the target industry; if the relevance of a satisfaction survey question is greater than 60% but less than 80%, then the satisfaction survey question will be used as a quasi-satisfaction survey question for the target industry; If the relevance of a satisfaction survey question is less than 60%, then the satisfaction survey question will be used as an alternative satisfaction survey question for the target industry; When the industry relevance is between 80% and 100%, if the relevance of a satisfaction survey question is greater than 70%, then the satisfaction survey question will be used as the satisfaction survey question for the target industry; If the relevance of a satisfaction survey question is greater than 60% and less than 70%, then the satisfaction survey question will be used as a quasi-satisfaction survey question for the target industry; If the relevance of a satisfaction survey question is less than 60%, the satisfaction survey question will be used as an alternative satisfaction survey question for the target industry.

9. A system for automatically processing satisfaction data based on big data resources, comprising a processor storing a computer executable program, characterized in that: The processor processes the satisfaction data according to the automatic processing method of satisfaction data based on big data resources as described in any one of claims 1-8.

Images