Intelligent qualitative data analysis method and system based on AIGC large model

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an intelligent qualitative data analysis method based on an AIGC large model, and the method comprises the following steps: S1, constructing a multi-dimensional scoring model, and automatically screening participants according with a qualitative research project; s2, intelligently generating an interview outline matched with the current qualitative research project by utilizing historical data of the qualitative research project and combining a model fine tuning technology and a retrieval enhancement algorithm; s3, monitoring the interview progress in real time through an AIGC large model, recording interview record data, transversely comparing the interview record data with the interview outline generated in S2, generating a guide problem and a guide prompt of the current interview progress, and guiding a host to interview; s4, deeply mining the interview record data to obtain a data analysis result; s5, the AIGC large model automatically generates a personalized research report according to a data analysis result and report requirements; and S6, multi-layer encryption and access control technologies are adopted to ensure the privacy and security of the data in the acquisition, processing, storage and transmission processes.

Description

Technical Field

[0001] The present invention relates to the application of artificial intelligence in the field of qualitative data research and analysis, and in particular to an intelligent qualitative data analysis method and system based on the AIGC big model, aiming to improve the efficiency and accuracy of qualitative data analysis. Background Art

[0002] In fields such as consumer research and user experience research, qualitative data analysis is a key means of tapping into users' deep cognition. Traditional qualitative research methods such as focus group discussions (FGD) and in-depth interviews (IDI) can collect multifaceted views on products, concepts, or hypotheses, but there are many challenges. In particular, the process of identifying and screening effective participants is complicated, research discussions are highly dependent on the level of moderators and interviewers, emotion and attitude identification is highly subjective, and data recording and analysis are time-consuming and laborious. These limitations lead to the instability of research results and, to a certain extent, limit the breadth and depth of research.

[0003] With the rapid development of natural language processing, pattern recognition and machine learning technologies, especially the rise of large language models, the development of intelligent qualitative data analysis methods and systems based on artificial intelligence generated content (AIGC) large models has become particularly necessary. Large language models have shown strong capabilities in understanding context, generating language and identifying emotions, providing a new opportunity to break through the bottleneck in traditional qualitative research methods.

[0004] This application proposes an innovative intelligent qualitative data analysis system that makes full use of the latest artificial intelligence technology. The system can automatically identify participants, generate interview outlines, host the interview process, and generate personalized reports through in-depth analysis of qualitative data. The system uses advanced natural language processing algorithms to analyze feedback and sentiment analysis technology to identify emotions and attitudes, thereby generating accurate and efficient reports. Summary of the invention

[0005] The present invention provides an intelligent qualitative data analysis method and system based on the AIGC large model to solve the problems raised in the above background technology.

[0006] To achieve the above objectives, in a first aspect, the present invention provides the following technical solutions:

[0007] An intelligent qualitative data analysis method based on the AIGC large model includes the following steps:

[0008] S1: Construct a multidimensional scoring model to automatically screen participants who meet the qualitative research project requirements;

[0009] S2: Using historical data from qualitative research projects, combined with model fine-tuning techniques and retrieval enhancement algorithms, intelligently generate interview outlines that match the current qualitative research project;

[0010] S3: Use the AIGC big model to monitor the progress of the interview in real time, record the interview transcript data, compare it horizontally with the interview outline generated by S2, generate guiding questions and guidance prompts for the current interview process, and guide the host to conduct the interview;

[0011] S4: Set multi-angle analysis prompts, conduct in-depth mining of interview transcript data, and obtain data analysis results;

[0012] S5: The AIGC big model automatically generates personalized research reports based on data analysis results and report requirements;

[0013] S6: Use multi-layer encryption and access control technology to ensure the privacy and security of data during collection, processing, storage and transmission.

[0014] Preferably, step S1 specifically includes:

[0015] S11. Obtaining original information of candidate participants from multiple data sources and performing data cleaning and formatting;

[0016] S12, extracting candidate participant characteristics and project characteristics from the data obtained in S11 using the AIGC big model;

[0017] S13. Construct a multidimensional scoring model and use the historical data of qualitative research projects and candidate participants to verify and optimize the multidimensional scoring model;

[0018] S14. Use the constructed multidimensional scoring model to screen participants who meet the needs of the qualitative research project.

[0019] Preferably, the construction of the multi-dimensional scoring model in step S13 includes:

[0020] (1) Construct multiple scoring items corresponding to the characteristics of candidate participants:

[0021] Age matching S age : Calculate the matching degree S between the candidate participant age and the age range of the qualitative research project focus group age :

[0022]

[0023] Where: x is the age of the candidate participant;

[0024] Gender matching S gender : Calculate the match S between the gender of the candidate participants and the gender of the focus group of the qualitative research project gender:

[0025]

[0026] Gender candidate Gender is the gender of the candidate participant. project Gender of the target group;

[0027] Education level matching S education : Calculate the matching degree S between the candidate participants’ education level and the education level of the qualitative research project’s focus group education :

[0028]

[0029] Among them: Education candidate Education level of candidate participants;

[0030] Geographic location matching S geo : Based on the degree of geographical match between the candidate participant’s place of residence and the group of interest of the qualitative research project S geo :

[0031]

[0032] Where: d is the distance between the candidate participant’s residence and the geographic location of the focus group of the qualitative research project;

[0033] Occupational relevance job :Based on the similarity between the candidate participants’ occupations and the occupations of the focus group of the qualitative research project, the occupational feature vector V job The cosine similarity is used to calculate the occupational relevance S job :

[0034]

[0035] Among them: job1 is the occupation of the candidate participants, job2 is the occupation of the focus group of the qualitative research project;

[0036] Interest matching degree S interest :Based on the similarity between the interests of candidate participants and the content of the qualitative research project, the interest feature vector V interest The cosine similarity of interest matching S is calculated interest :

[0037]

[0038] Among them: interest1 is the hobbies of the candidate participants, interest2 is the project content of the qualitative research project;

[0039] Socioeconomic status ses :Based on the different dimensions of the candidate participants’ income and occupational status in different regions, normalization is performed and the value range is [0, 1], S ses is the average of the two normalized values;

[0040] Potential participation S exp Calculate the potential participation S based on the number and quality of the candidate participants’ past participation in qualitative research projects exp :

[0041] S exp =num_projects×avg_rating

[0042] Among them: num_projects is the number of qualitative research projects that the candidate participants have participated in, and avg_rating is the average evaluation score of the qualitative research projects they have participated in;

[0043] (2) Assign weights and calculate comprehensive scores: Set the weights w for each scoring item. i , calculate the comprehensive score for each candidate participant S i is the score of each scoring item;

[0044] Step S14 specifically includes:

[0045] The multi-dimensional scoring model ranks the comprehensive scores of candidate participants from high to low, gives priority to candidate participants with high comprehensive scores, and generates a detailed report on the candidate participants based on the AIGC big model.

[0046] Preferably, step S2 specifically includes:

[0047] S21. Use historical data from qualitative research projects to build a knowledge base;

[0048] S22, build a multi-domain training dataset based on the knowledge base built in S21;

[0049] S23, fine-tune the pre-trained model using the multi-domain training dataset constructed in S22;

[0050] S24. Generate an interview outline based on the research tasks of the qualitative research project;

[0051] S25. Provide supplements for generating interview outlines by continuously improving and expanding the knowledge base.

[0052] Preferably, step S23 specifically includes:

[0053] (1) Model selection: Comprehensively evaluate the generation quality, computing resource requirements, and response speed of the open source AIGC large model, and select a pre-training model that has strong language generation capabilities and meets actual application conditions;

[0054] (2) Fine-tuning process: Use the constructed multi-domain training dataset to fine-tune the pre-trained model, set training parameters, ensure that the pre-trained model gradually adapts to specific tasks, and use Tokenizers to convert text data into token sequences for pre-trained model input for training;

[0055] (3) Performance evaluation: Use BLEU scores or GPT-4 model scoring to measure the accuracy and relevance of the content generated by the fine-tuned pre-trained model in different fields, and perform performance evaluation on the fine-tuned model;

[0056] 1) The calculation of the BLEU score includes:

[0057] a. Calculate n-gram precision P n : For each n, calculate the n-gram matches between the generated text and the reference text:

[0058]

[0059] Where: n is the number of characters (or words), which refers to the number of Chinese characters in Chinese;

[0060] An n-gram is a sequence of n characters (or words) that appear consecutively in a text, that is, a combination of n Chinese characters that appear consecutively in Chinese.

[0061] Count(ngram) is assigned the number of occurrences of n-gram in the generated text;

[0062] Count clip (ngram) is assigned a value based on the number of occurrences of the n-gram in the generated text, but the value assigned does not exceed the number of occurrences of the n-gram in the reference text;

[0063] Count ref (ngram) is assigned by the number of occurrences of n-gram in the reference text;

[0064] b. Calculate the geometric mean precision P: Take the geometric mean of the n-gram precisions for different n values:

[0065]

[0066] Where: N is the length of the largest n-gram;

[0067] c. Calculate the length penalty BP: The length penalty BP is used to penalize the case where the generated text is too short:

[0068]

[0069] where c is the length of the generated text and r is the length of the reference text;

[0070] d. Calculate the BLEU score:

[0071] BLEU = BP · P

[0072] e. Model performance judgment:

[0073] Preset the first performance compliance threshold as E. If BLEU ≥ E, it is judged that the performance of the pre-trained model meets the standard;

[0074] If BLEU < E, it is judged that the performance of the pre-trained model does not meet the standard;

[0075] 2) Using the GPT-4 model for scoring specifically includes:

[0076] a. Content similarity X: Analyze the similarity in content between the generated interview outline and the reference text, with the scoring range from 0 to 1;

[0077] b. Semantic consistency Y: Analyze the quality of the generated interview outline text in terms of logical order and semantic coherence, with the scoring range from 0 to 1;

[0078] c. Detail richness Z: Analyze the specificity, detail coverage, and clarity of the problem setting of the generated interview outline, with the scoring range from 0 to 1;

[0079] d. Calculate the GPT-4 score F: Perform weighted calculation on the content similarity X, semantic consistency Y, and detail richness Z to calculate the GPT-4 score F:

[0080] F = f 1 · X + f 2 · Y + f 3 · Z

[0081] where: f 1 、f 2 、f 3 are weight coefficients;

[0082] e. Model performance judgment:

[0083] Preset the second performance compliance threshold as H. If F ≥ H, it is judged that the performance of the pre-trained model meets the standard;

[0084] If F < H, it is judged that the performance of the pre-trained model does not meet the standard;

[0085] (4) Adjustment and optimization: If the BLEU score or GPT-4 score F of the pre-trained model does not reach the expected threshold, necessary model adjustments and optimizations will be performed. Adjustment measures include adjusting the learning rate, increasing training data, and optimizing the model architecture.

[0086] Preferably, step S3 comprises:

[0087] S31. Build a customized prompt word library containing multiple topics and analysis functions for application scenarios to guide the AIGC big model to accurately identify key information and topics in real-time interview records;

[0088] S32. Use the AIGC big model to dynamically analyze and extract key points from real-time interview records;

[0089] S33, AIGC large model compares and analyzes the key points extracted by S32 with the interview outline generated by S2 to detect any possible deviations in a timely manner;

[0090] S34. Combining the prompt word technology and the results of comparative analysis, the AIGC big model is used to generate guiding questions and guiding prompts suitable for the current interview process in real time.

[0091] Preferably, step S4 specifically includes:

[0092] S41. Set a series of segmentation analysis prompts, covering consumer portrait analysis, point extraction and analysis, product concept evaluation and user experience;

[0093] S42. Apply natural language processing technology to extract key data from interview transcripts and build detailed consumer portraits;

[0094] S43. Use sentiment analysis and topic modeling techniques to extract users’ key opinions and sentiment tendencies from interview transcript data, and generate detailed summaries and key point presentations;

[0095] S44. Use semantic analysis technology to simulate user feedback and evaluate the acceptance and market potential of product concepts;

[0096] S45. Combine user journey mapping technology to generate a user journey storyboard, comprehensively analyze the interaction process between users and products or services, and identify key moments in the experience;

[0097] S46. Introduce interactive analysis functions to support researchers in conducting supplementary analysis.

[0098] Preferably, step S5 specifically includes:

[0099] S51. Researchers set the report format and structure according to their needs;

[0100] S52, AIGC big model generates detailed and personalized research reports based on data analysis results and reporting requirements.

[0101] Preferably, step S6 specifically includes:

[0102] S61. During data collection and transmission, use the Advanced Encryption Standard to encrypt data;

[0103] S62. Use role-based access control mechanisms to ensure that only authorized personnel can access sensitive data;

[0104] S63. Anonymize the personal identity information of participants to protect privacy;

[0105] S64. Implement logging and monitoring mechanisms to regularly check the security status of the system and data access records;

[0106] S65. All processing models and data analysis processes are performed in a local privately deployed computing environment to ensure that data will not be leaked.

[0107] In a second aspect, the present invention provides an intelligent qualitative data analysis system based on the AIGC big model, which is applicable to the intelligent qualitative data analysis method based on the AIGC big model as described above, including:

[0108] Automatic identification module: including data collection unit, feature extraction unit, scoring system construction unit and application unit, automatically identifying and screening project participants;

[0109] Intelligent interview outline generation module: including knowledge base construction unit, training data set construction unit, model fine-tuning unit and interview outline generation unit. It uses historical data of qualitative research projects, combined with model fine-tuning technology and retrieval enhancement algorithm to automatically generate interview outlines.

[0110] Intelligent host assistant module: including prompt word library construction unit, comparative analysis unit and guiding question generation unit, which analyzes the real-time interview records based on the AIGC big model and interview outline, generates guiding questions and prompts, and assists the host in guiding the interview;

[0111] Transcript analysis module: including analysis prompt word setting unit, consumer portrait construction unit, viewpoint and key point extraction unit, product concept evaluation unit, user experience analysis unit and interactive analysis unit, which conducts in-depth mining of interview transcript data and obtains data analysis results;

[0112] Personalized report generation module: It includes a template setting unit and a data-driven generation unit. Based on data analysis results and report requirements, it automatically generates personalized qualitative research reports through natural language generation technology;

[0113] Data privacy protection module: includes data encryption unit, access control unit, data anonymity unit, full audit unit and local private deployment environment to ensure the privacy and security of data during collection, processing, storage and transmission.

[0114] Compared with the prior art, the present invention has the following beneficial effects:

[0115] The introduction of artificial intelligence technology aims to optimize the entire process of qualitative data research and analysis, including participant identification, data collection, interview outline generation and hosting, data storage and processing, in-depth data analysis, and report generation. By using AIGC-based large model technology, the efficiency and accuracy of qualitative data analysis can be significantly improved, and subjective bias and resource consumption in the process of manual intervention can be reduced. This not only provides higher quality services for consumer research and user experience research, but also expands the application scope of qualitative data analysis to meet a wider range of market needs. The system uses the latest artificial intelligence technology to significantly improve the efficiency and accuracy of qualitative data analysis, while reducing costs and improving data reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0116] Figure 1 A schematic diagram of a process for automatically screening participants who meet the requirements of a qualitative research project in an intelligent qualitative data analysis method based on an AIGC large model provided in an embodiment of the present invention;

[0117] Figure 2 A schematic diagram of a process for automatically generating an interview outline in an intelligent qualitative data analysis method based on an AIGC large model provided in an embodiment of the present invention;

[0118] Figure 3 A schematic diagram of a flow chart of generating guiding questions and guiding prompts in an intelligent qualitative data analysis method based on an AIGC large model provided in an embodiment of the present invention;

[0119] Figure 4 A structural block diagram of an intelligent qualitative data analysis system based on the AIGC large model provided for an implementation manner of the present invention. DETAILED DESCRIPTION

[0120] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0121] The embodiment of the present invention provides an intelligent qualitative data analysis method based on the AIGC large model, comprising the following steps:

[0122] S1: Construct a multidimensional scoring model to automatically screen participants who meet the qualitative research project requirements;

[0123] S2: Using historical data from qualitative research projects, combined with model fine-tuning techniques and retrieval enhancement algorithms, intelligently generate interview outlines that match the current qualitative research project;

[0124] S3: Use the AIGC big model to monitor the progress of the interview in real time, record the interview transcript data, compare it horizontally with the interview outline generated by S2, generate guiding questions and guidance prompts for the current interview process, and guide the host to conduct the interview;

[0125] S4: Set multi-angle analysis prompts, conduct in-depth mining of interview transcript data, and obtain data analysis results;

[0126] S5: The AIGC big model automatically generates personalized research reports based on data analysis results and report requirements;

[0127] S6: Use multi-layer encryption and access control technology to ensure the privacy and security of data during collection, processing, storage and transmission.

[0128] The intelligent qualitative data analysis method based on the AIGC big model provided by the present invention introduces artificial intelligence technology to optimize the entire process of qualitative data research and analysis, including participant screening, data collection, interview outline generation and hosting, data storage and processing, in-depth data analysis, and report generation. By using the AIGC big model technology, the efficiency and accuracy of qualitative data analysis can be significantly improved, and the subjective bias and resource consumption in the manual intervention process can be reduced. This not only provides higher quality services for consumer research and user experience research, but also expands the application scope of qualitative data analysis to meet a wider range of market needs.

[0129] Figure 1 A flowchart of automatically screening participants who meet the requirements of a qualitative research project in an intelligent qualitative data analysis method based on the AIGC large model provided by the embodiment of the present invention. Figure 1 As shown, in one embodiment of the present invention, step S1 specifically includes:

[0130] S11. Obtaining original information of candidate participants from multiple data sources and performing data cleaning and formatting;

[0131] S12, extracting candidate participant characteristics and project characteristics from the data obtained in S11 using the AIGC big model;

[0132] S13. Construct a multidimensional scoring model and use the historical data of qualitative research projects and candidate participants to verify and optimize the multidimensional scoring model;

[0133] S14. Use the constructed multidimensional scoring model to screen participants who meet the needs of the qualitative research project.

[0134] Specifically, step S11 includes:

[0135] (1) Data acquisition: The original information of candidate participants is obtained from multiple data sources through data interfaces. The multiple data sources include but are not limited to social media platforms, online forums, questionnaire databases, and private sample libraries. These data sources can be provided in the form of API interfaces, data files, etc.

[0136] (2) Data preprocessing: First, data cleaning is performed to remove invalid, erroneous, and incomplete records, handle missing values, and identify and delete duplicates. Then, data redundancy and inconsistency are avoided by merging data from different sources. Missing qualitative data are recorded as “unknown” and treated as 0 in subsequent scoring calculations.

[0137] Specifically, step S12 includes:

[0138] (1) Extract characteristics of candidate participants: including age, gender, education level, geographic location, occupation, hobbies, socioeconomic status, and previous experience in participating in research projects;

[0139] (2) Extract project features: including project goals, focus groups, project content, and keywords.

[0140] Project goals: clearly define the specific goals of the qualitative research project; a. For example, improving product satisfaction, increasing user engagement, improving service quality, etc.;

[0141] Focus group: clearly define the target audience of the qualitative research project, including age range (e.g., 25-35 years old), gender (e.g., female), education level (e.g., bachelor degree and above), occupation (e.g., students, white-collar workers, etc.), and geographic location (specific city or region);

[0142] Project content: clearly define the core theme of the qualitative research project, such as electronic products, healthy lifestyle, etc., as well as specific content, such as electronic product user experience feedback, healthy diet survey;

[0143] Keywords: Get multiple keywords based on the project content, such as "satisfaction survey", "product feedback", "user experience", etc.

[0144] (3) Text feature vectorization: The extracted candidate participant features and project features are vectorized to facilitate the subsequent calculation of the scoring item scores. Pre-trained embedding models (such as BERT) can be used to achieve vectorization.

[0145] Specifically, the construction of the multi-dimensional scoring model in step S13 includes:

[0146] (1) Construct multiple scoring items corresponding to the characteristics of candidate participants:

[0147] Age matching S age : Calculate the matching degree S between the candidate participant age and the age range of the qualitative research project focus group age :

[0148]

[0149] Where: x is the age of the candidate participant;

[0150] Gender matching S gender : Calculate the match S between the gender of the candidate participants and the gender of the focus group of the qualitative research project gender :

[0151]

[0152] Gender candidate Gender is the gender of the candidate participant. project Gender of the target group;

[0153] Education level matching S education : Calculate the matching degree S between the candidate participants’ education level and the education level of the qualitative research project’s focus group education :

[0154]

[0155] Among them: Education candidate Education level of candidate participants;

[0156] Geographic location matching S geo : Based on the degree of geographical match between the candidate participant’s place of residence and the group of interest of the qualitative research project S geo :

[0157]

[0158] Where: d is the distance between the candidate participant’s residence and the geographic location of the focus group of the qualitative research project;

[0159] Occupational relevance job :Based on the similarity between the candidate participants’ occupations and the occupations of the focus group of the qualitative research project, the occupational feature vector V job The cosine similarity is used to calculate the occupational relevance S job :

[0160]

[0161] Among them: job1 is the occupation of the candidate participants, job2 is the occupation of the focus group of the qualitative research project;

[0162] Interest matching degree S interest :Based on the similarity between the interests of candidate participants and the content of the qualitative research project, the interest feature vector V interest The cosine similarity of interest matching S is calculated interest :

[0163]

[0164] Among them: interest1 is the hobbies of the candidate participants, interest2 is the project content of the qualitative research project;

[0165] Socioeconomic status ses :Based on the different dimensions of the candidate participants’ income and occupational status in different regions, normalization is performed and the value range is [0, 1], S ses is the average of the two normalized values;

[0166] Potential participation S exp Calculate the potential participation S based on the number and quality of the candidate participants’ past participation in qualitative research projects exp :

[0167] S exp =num_projects×avg_rating

[0168] Among them: num_projects is the number of qualitative research projects that the candidate participants have participated in, and avg_rating is the average evaluation score of the qualitative research projects they have participated in;

[0169] (2) Assign weights and calculate comprehensive scores: Set the weights w for each scoring item. i , calculate the comprehensive score for each candidate participant S i is the score of each rating item.

[0170] Specifically, the verification and optimization of the multi-dimensional scoring model using the historical data of the qualitative research project and the historical data of the candidate participants described in step S13 includes:

[0171] Adjust the calculation method and weight of each scoring item based on the verification results to improve the accuracy and rationality of the multi-dimensional scoring model.

[0172] Specifically, step S14 includes:

[0173] The multi-dimensional scoring model ranks the comprehensive scores of candidate participants from high to low, gives priority to candidate participants with high comprehensive scores, and generates a detailed report on candidate participants based on the AIGC big model; the detailed report contains a detailed description of the candidate participant characteristics and scoring criteria for each candidate participant, as well as reminders about missing and outdated characteristics.

[0174] Figure 2 A flowchart of automatically generating an interview outline in an intelligent qualitative data analysis method based on the AIGC big model provided by the embodiment of the present invention. Figure 2 As shown, in one embodiment of the present invention, step S2 specifically includes:

[0175] S21. Use historical data from qualitative research projects to build a knowledge base;

[0176] S22, build a multi-domain training dataset based on the knowledge base built in S21;

[0177] S23, fine-tune the pre-trained model using the multi-domain training dataset constructed in S22;

[0178] S24. Generate an interview outline based on the research tasks of the qualitative research project;

[0179] S25. Provide supplements for generating interview outlines by continuously improving and expanding the knowledge base.

[0180] Specifically, step S21 includes:

[0181] (1) Data collection: Extract project background data and interview outlines from historical data of qualitative research projects, covering multiple fields and various qualitative research objectives; such as product testing, consumer behavior research, user experience analysis, new product co-creation research, consumer demand research, etc. Data sources include qualitative research reports, interview records, questionnaire results, etc.

[0182] (2) Organize data: classify, organize and label the extracted data; use specific fields such as research background (such as xx product testing project), research purpose (such as understanding consumer attitudes, product cognition and behavioral habits, purchase decision-making process, etc.), research object (specific consumer group), qualitative discussion outline (specific discussion process and questions) and other project-related data (such as project number, execution date, etc.) to classify and label the data, so as to facilitate accurate classification and efficient management of the data;

[0183] (3) Build a knowledge base: Import the sorted data into unstructured search tools, create indexes for efficient retrieval, and form a systematic knowledge base. For example, for text data, create a full-text keyword index in Elasticsearch and use Milvus for vectorization to achieve efficient semantic retrieval.

[0184] Specifically, step S22 includes:

[0185] (1) Select relevant data: Select representative interview outlines and research objectives covering multiple fields from the knowledge base to form a preliminary multi-field training data set; for example, select interview outlines of typical consumer decision-making processes from a consumer behavior research project;

[0186] (2) Data cleaning: remove incomplete, duplicate, and meaningless parts of the data, deduplicate, proofread, and format the text, use natural language processing technology to correct spelling errors, and unify the format;

[0187] (3) Data formatting: Format the cleaned data into a structure suitable for model training. For example, each piece of data includes an input query (research objective: such as "xx product concept test") and an output template (interview outline: such as "1. Overall impression; 2. Whether the concept is credible; 3. Suggestions for improvement, etc."), in the JSON structure.

[0188] Specifically, step S23 includes:

[0189] (1) Model selection: Comprehensively evaluate the generation quality, computing resource requirements, and response speed of the open source AIGC large model, and select the model with strong language generation capabilities and that meets actual application conditions as the pre-training model;

[0190] (2) Fine-tuning process: Use the constructed multi-domain training dataset to fine-tune the pre-trained model, set training parameters, ensure that the pre-trained model gradually adapts to specific tasks, and use Tokenizers to convert text data into token sequences for pre-trained model input for training. For example, training parameters can include learning rate 0.001, batch size 32, training cycle 10 times, etc. The fine-tuning process should include multiple rounds of verification and optimization, and avoid overfitting through early stopping mechanism;

[0191] (3) Performance evaluation: Use BLEU scores or GPT-4 model scoring to measure the accuracy and relevance of the content generated by the fine-tuned pre-trained model in different fields, and evaluate the performance of the fine-tuned model; a high BLEU score indicates that the generated text is highly similar to the reference text in terms of words, phrases, and overall content, which meets the requirements of qualitative research; at the same time, the generated text and reference text are input into the GPT-4 model, and its powerful natural language processing capabilities are used for a more detailed and comprehensive evaluation; the GPT-4 model can provide in-depth evaluation results through understanding and reasoning of the text;

[0192] 1) The calculation of the BLEU score includes:

[0193] a. Calculate n-gram precision P n : For each n, calculate the n-gram matches between the generated text and the reference text:

[0194]

[0195] Where: n is the number of characters (or words), which refers to the number of Chinese characters in Chinese;

[0196] An n-gram is a sequence of n characters (or words) that appear consecutively in a text, that is, a combination of n Chinese characters that appear consecutively in Chinese. In the embodiment of the present invention, n is 2-4 according to the usage habits of Chinese.

[0197] Count(ngram) is assigned the number of occurrences of n-gram in the generated text;

[0198] Count clip (ngram) is assigned a value based on the number of occurrences of the n-gram in the generated text, but the value assigned does not exceed the number of occurrences of the n-gram in the reference text;

[0199] Count ref (ngram) is assigned by the number of occurrences of n-gram in the reference text;

[0200] b. Calculate the geometric mean precision P: Take the geometric mean of the n-gram precisions for different n values:

[0201]

[0202] Wherein: N is the maximum length of n-gram. In the embodiment of the present invention, according to the usage habits of Chinese, N is 4;

[0203] c. Calculate the length penalty BP: The length penalty BP is used to penalize the situation where the generated text is too short:

[0204]

[0205] Where: c is the length of the generated text, r is the length of the reference text;

[0206] d. Calculate BLEU score:

[0207] BLEU=BP·P

[0208] e. Model performance judgment:

[0209] The first performance threshold is preset to E. If BLEU ≥ E, the performance of the pre-trained model is judged to be up to standard.

[0210] If BLEU < E, the performance of the pre-trained model is judged to be substandard;

[0211] In a preferred embodiment of the present invention, E = 0.45;

[0212] 2) The GPT-4 model scoring includes:

[0213] a. Content similarity X: The similarity between the generated interview outline and the reference text in terms of content, with a score ranging from 0 to 1;

[0214] b. Semantic consistency Y: Analyze the quality of the generated interview outline text in terms of logical order and semantic coherence, with a score ranging from 0 to 1;

[0215] c. Detail Richness Z: Analyze the specificity of the generated interview outline, the degree of detail coverage, and the clarity of the question setting, with a score range of 0 to 1;

[0216] d. Calculate the GPT-4 score F: Weight the content similarity X, semantic consistency Y, and detail richness Z to calculate the GPT-4 score F:

[0217] F=f 1 ·X+f 2 ·Y+f 3 ·Z

[0218] Where: f 1 、f 2 、f 3 is the weight coefficient; in a preferred embodiment of the present invention, f 1 =0.4, f 1 =0.3, f 1 =0.3;

[0219] e. Model performance judgment:

[0220] The second performance threshold is preset to H. If F ≥ H, it is determined that the performance of the pre-trained model meets the standard.

[0221] If F<H, the performance of the pre-trained model is judged to be substandard;

[0222] In a preferred embodiment of the present invention, F = 0.7;

[0223] (4) Adjustment and optimization: If the BLEU score or GPT-4 score F of the pre-trained model does not reach the expected threshold, necessary model adjustments and optimizations will be performed. Adjustment measures include adjusting the learning rate, increasing training data, and optimizing the model architecture.

[0224] Specifically, step S24 includes:

[0225] (1) Retrieve relevant information: Use the search enhancement algorithm to retrieve qualitative research objectives and interview outlines that are highly relevant to the current qualitative research task from the knowledge base; the search enhancement algorithm comprehensively considers keyword matching and semantic similarity to ensure the accuracy and relevance of the search results;

[0226] (2) Generate interview outline: The retrieved information is input into the fine-tuned pre-trained model, which combines the contextual information to generate an interview outline that meets the needs of the current qualitative research project. For example, if the user inputs "formulate an outline for user experience research on product xx", the knowledge base first retrieves relevant historical data, and then combines the fine-tuned pre-trained model to output a complete interview outline.

[0227] Specifically, step S25 includes:

[0228] We will continue to improve and expand our knowledge base by leveraging the gradually accumulated qualitative research projects in more fields and directions, supplement the generation of interview outlines, and improve their accuracy, diversity, and relevance. At the same time, we will regularly fine-tune the pre-trained model to adapt to new qualitative research needs and data changes.

[0229] Figure 3 A flowchart of generating guiding questions and guiding prompts in an intelligent qualitative data analysis method based on an AIGC large model provided by an embodiment of the present invention. Figure 3 As shown, in one embodiment of the present invention, step S3 includes:

[0230] S31. Build a customized prompt word library containing multiple topics and analysis functions for application scenarios to guide the AIGC big model to accurately identify key information and topics in real-time interview records;

[0231] S32. Use the AIGC big model to dynamically analyze and extract key points from real-time interview records;

[0232] S33, AIGC large model compares and analyzes the key points extracted by S32 with the interview outline generated by S2 to detect any possible deviations in a timely manner;

[0233] S34. Combining the prompt word technology and the results of comparative analysis, the AIGC big model is used to generate guiding questions and guiding prompts suitable for the current interview process in real time.

[0234] Specifically, step S31 includes:

[0235] (1) Pre-set prompts: Based on the interview outline and research objectives, design a set of prompts for analysis and to assist moderators. Prompts should be based on domain-specific knowledge and research focus, covering a variety of topics, keywords, and phrases. The design process should consider multiple factors such as semantic similarity, contextual relevance, and topic coverage.

[0236] (2) Constructing a prompt word library: Based on the designed prompt words, a multi-dimensional, multi-domain preset prompt word library is constructed and structured to optimize query efficiency. The word library combines natural language processing technology and contains common terms, key topics, and specific phrases in the field to accurately identify and extract important information in the interview.

[0237] (3) Optimize the prompt word library: The prompt word library needs to be updated and maintained regularly to ensure its effectiveness and compatibility in different interview scenarios.

[0238] Specifically, step S32 includes:

[0239] (1) Real-time analysis of interview records: The AIGC model analyzes the content of interview conversations in real time, extracts key information, identifies key topics, important keywords, and core expressions, and generates preliminary analysis results;

[0240] (2) Generate effective prompts: Based on the preliminary analysis results, further classify and summarize the keywords, themes, and phrases in the interviews and generate effective prompts. For example, when the interviewees often mention the "practicality" of the new features, the system will mark "practicality" as important information.

[0241] (3) Organize information: The system formats and archives the extracted key information to ensure that the key information is clearly organized and of reliable quality, facilitating subsequent in-depth analysis and application.

[0242] Specifically, step S33 includes:

[0243] (1) Real-time comparison of interview outlines: The AIGC model compares the extracted key points with the pre-set interview outline to ensure that the interview content is consistent with the research plan and to promptly detect deviations or omissions. The comparison is based on intelligent dialogue analysis technology to ensure that every detail is accurately proofread.

[0244] (2) Generate a gap report: If the AIGC model finds that the preset questions in the interview outline are not discussed, or that some topics in the interview outline need to be discussed in depth, the AIGC model will automatically generate a gap report. For example, if the preset question in the interview outline, "What do you think about the battery life of the new product?" is not mentioned, the AIGC model will record this in the gap report.

[0245] Specifically, step S34 includes:

[0246] (1) Generate guiding questions and prompts: The AIGC big model combines the conclusions of the gap report and the current discussion content, and uses natural language generation technology to generate relevant guiding questions and prompts. For example, guiding questions such as "In what aspects do you think this new product needs to be improved?" will help the host to dig deeper into the participants' insights;

[0247] (2) Providing guiding questions and guiding prompts: The AIGC big model will automatically and in real time display the generated guiding questions and guiding prompts to the host, allowing the host to flexibly adjust the direction of the interview.

[0248] In one embodiment of the present invention, step S4 specifically includes:

[0249] S41. Set a series of segmentation analysis prompts, covering consumer portrait analysis, point extraction and analysis, product concept evaluation and user experience; these prompts should include but are not limited to: consumer interests, occupations, gender, age groups, hobbies, pain points, usage scenarios, etc.; use these prompts to systematically extract and organize relevant information for efficient in-depth analysis;

[0250] S42. Apply natural language processing technology to extract key data from interview transcripts and build detailed consumer portraits. This process focuses on identifying consumer behavior patterns, preferences, and needs. Through refined text analysis, key features such as consumer purchase motivations, decision-making processes, usage frequency, and satisfaction are recorded in detail to form structured data reports.

[0251] S43. Use sentiment analysis and topic modeling techniques to extract users’ key opinions and sentiment tendencies from interview transcript data and generate detailed summaries and key point presentations; including using sentiment analysis tools to identify positive, negative, and neutral emotions, using topic modeling (such as LDA) to identify frequently occurring topics in interviews, and generating visual charts or reports to facilitate stakeholders to quickly understand user feedback;

[0252] S44. Use semantic analysis technology to simulate user feedback and evaluate the acceptance and market potential of product concepts; use semantic similarity calculation methods to analyze user responses to different product concepts, predict user acceptance and improvement suggestions, generate detailed operational reports, and provide data support for product optimization;

[0253] S45. Combine user journey mapping technology to generate a user journey storyboard, comprehensively analyze the interaction process between users and products or services, and identify key moments in the experience; through detailed user journey analysis, simulate the entire process of users from cognition, consideration, purchase to use and feedback, and focus on marking the user's pleasure points, pain points and key decision points; use visualization tools to display the user journey, so as to intuitively understand and solve problems in user experience;

[0254] S46. Introducing interactive analysis functions to support researchers in supplementary analysis. Through dynamic query and filtering functions, researchers can flexibly retrieve interview data, add definition tags and annotations, and enrich analysis results; support real-time data updates, view adjusted results at any time, and provide data visualization tools to help users understand analysis results; at the same time, through interactive Q&A functions, using the AIGC large model to analyze interview records, support researchers to ask specific questions and obtain detailed analysis answers.

[0255] In one embodiment of the present invention, step S5 specifically includes:

[0256] S51. Researchers set the report format and structure according to their needs;

[0257] S52, AIGC big model generates detailed and personalized research reports based on data analysis results and reporting requirements.

[0258] In one embodiment of the present invention, step S6 specifically includes:

[0259] S61. During data collection and transmission, use the Advanced Encryption Standard to encrypt data;

[0260] S62. Use role-based access control mechanisms to ensure that only authorized personnel can access sensitive data;

[0261] S63. Anonymize the personal identity information of participants to protect privacy;

[0262] S64. Implement logging and monitoring mechanisms to regularly check the security status of the system and data access records;

[0263] S65. All processing models and data analysis processes are performed in a local privately deployed computing environment to ensure that data will not be leaked.

[0264] Figure 4 The following is a structural block diagram of an intelligent qualitative data analysis system based on the AIGC large model provided by the embodiment of the present invention. Figure 4 As shown, the present invention also provides an intelligent qualitative data analysis system based on the AIGC large model, which is applicable to the intelligent qualitative data analysis method based on the AIGC large model as described above, including:

[0265] Automatic identification module: including data collection unit, feature extraction unit, scoring system construction unit and application unit, automatically identifying and screening project participants;

[0266] Intelligent interview outline generation module: including knowledge base construction unit, training data set construction unit, model fine-tuning unit and interview outline generation unit. It uses historical data of qualitative research projects, combined with model fine-tuning technology and retrieval enhancement algorithm to automatically generate interview outlines.

[0267] Intelligent host assistant module: including prompt word library construction unit, comparative analysis unit and guiding question generation unit, which analyzes the real-time interview records based on the AIGC big model and interview outline, generates guiding questions and prompts, and assists the host in guiding the interview;

[0268] Transcript analysis module: including analysis prompt word setting unit, consumer portrait construction unit, viewpoint and key point extraction unit, product concept evaluation unit, user experience analysis unit and interactive analysis unit, which conducts in-depth mining of interview transcript data and obtains data analysis results;

[0269] Personalized report generation module: It includes a template setting unit and a data-driven generation unit. Based on data analysis results and report requirements, it automatically generates personalized qualitative research reports through natural language generation technology;

[0270] Data privacy protection module: includes data encryption unit, access control unit, data anonymity unit, full audit unit and local private deployment environment to ensure the privacy and security of data during collection, processing, storage and transmission.

[0271] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent qualitative data analysis method based on the AIGC large model, characterized in that: The following steps are involved: S1: Construct a multidimensional scoring model to automatically screen participants who meet the qualitative research project requirements; S2: Using historical data from qualitative research projects, combined with model fine-tuning techniques and retrieval enhancement algorithms, intelligently generate interview outlines that match the current qualitative research project; S3: Use the AIGC big model to monitor the progress of the interview in real time, record the interview transcript data, compare it horizontally with the interview outline generated by S2, generate guiding questions and guidance prompts for the current interview process, and guide the host to conduct the interview; S4: Set multi-angle analysis prompts, conduct in-depth mining of interview transcript data, and obtain data analysis results; S5: The AIGC big model automatically generates personalized research reports based on data analysis results and report requirements; S6: Use multi-layer encryption and access control technology to ensure the privacy and security of data during collection, processing, storage and transmission.

2. The intelligent qualitative data analysis method based on the AIGC large model according to claim 1 is characterized in that: Step S1 specifically includes: S11. Obtaining original information of candidate participants from multiple data sources and performing data cleaning and formatting; S12, extracting candidate participant characteristics and project characteristics from the data obtained in S11 using the AIGC big model; S13. Construct a multidimensional scoring model and use the historical data of qualitative research projects and candidate participants to verify and optimize the multidimensional scoring model; S14. Use the constructed multidimensional scoring model to screen participants who meet the needs of the qualitative research project.

3. The intelligent qualitative data analysis method based on the AIGC large model according to claim 2 is characterized in that: The construction of the multi-dimensional scoring model in step S13 includes: (1) Construct multiple scoring items corresponding to the characteristics of candidate participants: Age matching S age : Calculate the matching degree S between the candidate participant age and the age range of the qualitative research project focus group age : Where: x is the age of the candidate participant; Gender matching S gender : Calculate the match S between the gender of the candidate participants and the gender of the focus group of the qualitative research project gender : Gender candidate Gender is the gender of the candidate participant. project Gender of the target group; Education level matching S education : Calculate the matching degree S between the candidate participants’ education level and the education level of the qualitative research project’s focus group education : Among them: Education candidate Education level of candidate participants; Geographic location matching S geo : Based on the degree of geographical match between the candidate participant’s place of residence and the group of interest of the qualitative research project S geo : Where: d is the distance between the candidate participant’s residence and the geographic location of the focus group of the qualitative research project; Occupational relevance job :Based on the similarity between the candidate participants’ occupations and the occupations of the focus group of the qualitative research project, the occupational feature vector V job The cosine similarity is used to calculate the occupational relevance S job : Among them: job1 is the occupation of the candidate participants, job2 is the occupation of the focus group of the qualitative research project; Interest matching degree S interest :Based on the similarity between the interests of candidate participants and the content of the qualitative research project, the interest feature vector V interest The cosine similarity of interest matching S is calculated interest : Among them: interest1 is the hobbies of the candidate participants, interest2 is the project content of the qualitative research project; Socioeconomic status ses :Based on the different dimensions of the candidate participants’ income and occupational status in different regions, normalization is performed and the value range is [0,1], S ses is the average of the two normalized values; Potential participation S exp Calculate the potential participation S based on the number and quality of the candidate participants’ past participation in qualitative research projects exp : S exp =num_projects×avg_rating Among them: num_projects is the number of qualitative research projects that the candidate participants have participated in, and avg_rating is the average evaluation score of the qualitative research projects they have participated in; (2) Assign weights and calculate comprehensive scores: Set the weights w for each scoring item. i , calculate the comprehensive score for each candidate participant S i is the score of each scoring item; Step S14 specifically includes: The multi-dimensional scoring model ranks the comprehensive scores of candidate participants from high to low, gives priority to candidate participants with high comprehensive scores, and generates a detailed report on the candidate participants based on the AIGC big model.

4. The intelligent qualitative data analysis method based on the AIGC large model according to claim 1 is characterized in that: Step S2 specifically includes: S21. Use historical data from qualitative research projects to build a knowledge base; S22, build a multi-domain training dataset based on the knowledge base built in S21; S23, fine-tune the pre-trained model using the multi-domain training dataset constructed in S22; S24. Generate an interview outline based on the research tasks of the qualitative research project; S25. Provide supplements for generating interview outlines by continuously improving and expanding the knowledge base.

5. The intelligent qualitative data analysis method based on the AIGC large model according to claim 4 is characterized in that: Step S23 specifically includes: (1) Model selection: Comprehensively evaluate the generation quality, computing resource requirements, and response speed of the open source AIGC large model, and select a pre-training model that has strong language generation capabilities and meets actual application conditions; (2) Fine-tuning process: Use the constructed multi-domain training dataset to fine-tune the pre-trained model, set training parameters, ensure that the pre-trained model gradually adapts to specific tasks, and use Tokenizers to convert text data into token sequences for pre-trained model input for training; (3) Performance evaluation: Use BLEU scores or GPT-4 model scoring to measure the accuracy and relevance of the content generated by the fine-tuned pre-trained model in different fields, and perform performance evaluation on the fine-tuned model; 1) The calculation of the BLEU score includes: a. Calculate n-gram precision P n : For each n, calculate the n-gram matches between the generated text and the reference text: Where: n: the number of characters (or words), in Chinese it refers to the number of Chinese characters; n-gram: a sequence of n characters (or words) that appear consecutively in a text, i.e., a combination of n consecutive Chinese characters in Chinese. Count(ngram) is assigned the number of occurrences of n-gram in the generated text; Count clip (ngram) is assigned a value based on the number of occurrences of the n-gram in the generated text, but the value assigned does not exceed the number of occurrences of the n-gram in the reference text; Count ref (ngram) is assigned by the number of occurrences of n-gram in the reference text; b. Calculate the geometric mean precision P: Take the geometric mean of the n-gram precisions for different n values: Where: N is the length of the largest n-gram; c. Calculate the length penalty BP: The length penalty BP is used to penalize the situation where the generated text is too short: Where: c is the length of the generated text, r is the length of the reference text; d. Calculate BLEU score: BLEU=BP·P e. Model performance judgment: The first performance threshold is preset to E. If BLEU ≥ E, the performance of the pre-trained model is judged to be up to standard. If BLEU < E, the performance of the pre-trained model is judged to be substandard; 2) The GPT-4 model scoring includes: a. Content similarity X: The similarity between the generated interview outline and the reference text in terms of content, with a score ranging from 0 to 1; b. Semantic consistency Y: Analyze the quality of the generated interview outline text in terms of logical order and semantic coherence, with a score ranging from 0 to 1; c. Detail Richness Z: Analyze the specificity of the generated interview outline, the degree of detail coverage, and the clarity of the question setting, with a score range of 0 to 1; d. Calculate the GPT-4 score F: Weight the content similarity X, semantic consistency Y, and detail richness Z to calculate the GPT-4 score F: F=f1·X+f2·Y+f3·Z Among them: f1, f2, f3 are weight coefficients; e. Model performance judgment: The second performance threshold is preset to H. If F ≥ H, it is determined that the performance of the pre-trained model meets the standard. If F<H, the performance of the pre-trained model is judged to be substandard; (4) Adjustment and optimization: If the BLEU score or GPT-4 score F of the pre-trained model does not reach the expected threshold, necessary model adjustments and optimizations will be performed. Adjustment measures include adjusting the learning rate, increasing training data, and optimizing the model architecture.

6. The intelligent qualitative data analysis method based on the AIGC large model according to claim 1 is characterized in that: Step S3 includes: S31. Build a customized prompt word library containing multiple topics and analysis functions for application scenarios to guide the AIGC big model to accurately identify key information and topics in real-time interview records; S32. Use the AIGC big model to dynamically analyze and extract key points from real-time interview records; S33, AIGC large model compares and analyzes the key points extracted by S32 with the interview outline generated by S2 to detect any possible deviations in a timely manner; S34. Combining the prompt word technology and the results of comparative analysis, the AIGC big model is used to generate guiding questions and guiding prompts suitable for the current interview process in real time.

7. The intelligent qualitative data analysis method based on the AIGC large model according to claim 1 is characterized in that: Step S4 specifically includes: S41. Set a series of segmentation analysis prompts, covering consumer portrait analysis, point extraction and analysis, product concept evaluation and user experience; S42. Apply natural language processing technology to extract key data from interview transcripts and build detailed consumer portraits; S43. Use sentiment analysis and topic modeling techniques to extract users’ key opinions and sentiment tendencies from interview transcript data, and generate detailed summaries and key point presentations; S44. Use semantic analysis technology to simulate user feedback and evaluate the acceptance and market potential of product concepts; S45. Combine user journey mapping technology to generate a user journey storyboard, comprehensively analyze the interaction process between users and products or services, and identify key moments in the experience; S46. Introduce interactive analysis functions to support researchers in conducting supplementary analysis.

8. The intelligent qualitative data analysis method based on the AIGC large model according to claim 1 is characterized in that: Step S5 specifically includes: S51. Researchers set the report format and structure according to their needs; S52, AIGC big model generates detailed and personalized research reports based on data analysis results and reporting requirements.

9. The intelligent qualitative data analysis method based on the AIGC large model according to claim 1 is characterized in that: Step S6 specifically includes: S61. During data collection and transmission, use the Advanced Encryption Standard to encrypt data; S62. Use role-based access control mechanisms to ensure that only authorized personnel can access sensitive data; S63. Anonymize the personal identity information of participants to protect privacy; S64. Implement logging and monitoring mechanisms to regularly check the security status of the system and data access records; S65. All processing models and data analysis processes are performed in a local privately deployed computing environment to ensure that data will not be leaked.

10. An intelligent qualitative data analysis system based on the AIGC large model, characterized in that: The intelligent qualitative data analysis method based on the AIGC large model according to any one of claims 1 to 9 comprises: Automatic identification module: including data collection unit, feature extraction unit, scoring system construction unit and application unit, automatically identifying and screening project participants; Intelligent interview outline generation module: including knowledge base construction unit, training data set construction unit, model fine-tuning unit and interview outline generation unit. It uses historical data of qualitative research projects, combined with model fine-tuning technology and retrieval enhancement algorithm to automatically generate interview outlines. Intelligent host assistant module: including prompt word library construction unit, comparative analysis unit and guiding question generation unit, which analyzes the real-time interview records based on the AIGC big model and interview outline, generates guiding questions and prompts, and assists the host in guiding the interview; Transcript analysis module: including analysis prompt word setting unit, consumer portrait construction unit, viewpoint and key point extraction unit, product concept evaluation unit, user experience analysis unit and interactive analysis unit, which conducts in-depth mining of interview transcript data and obtains data analysis results; Personalized report generation module: It includes a template setting unit and a data-driven generation unit. Based on data analysis results and report requirements, it automatically generates personalized qualitative research reports through natural language generation technology; Data privacy protection module: includes data encryption unit, access control unit, data anonymity unit, full audit unit and local private deployment environment to ensure the privacy and security of data during collection, processing, storage and transmission.

Images