Search

CN-122022904-A - Questionnaire personalized display method based on AI real-time analysis

CN122022904ACN 122022904 ACN122022904 ACN 122022904ACN-122022904-A

Abstract

The invention discloses a questionnaire personalized display method based on AI real-time analysis, which belongs to the technical field of crossing big data, artificial intelligence and market investigation, and comprises the following steps of S1, preparing questionnaires; S2, multi-source data acquisition, S3, real-time user image construction, S4, dynamic decision processing, and S5, personalized questionnaire display. According to the questionnaire personalized display method based on AI real-time analysis, through semantic rule setting, multi-source data real-time portrait construction and dynamic decision engine optimization, the complexity of questionnaire configuration is greatly reduced, flexibility is improved, real-time scenes of users are accurately adapted, the individuation degree of questionnaires is deepened, and investigation is more fit with actual situations of answering persons.

Inventors

  • HUANG PEI

Assignees

  • 上海众言网络科技有限公司

Dates

Publication Date
20260512
Application Date
20260126

Claims (9)

  1. 1. A questionnaire personalized display method based on AI real-time analysis is characterized by comprising the following steps: S1, editing a basic questionnaire title through a questionnaire setting module, setting logic requirements by adopting semantic matching degree, calling a lightweight model to analyze the requirements and generating an adaptation rule, and constructing a questionnaire logic rule base; s2, collecting answer behaviors, basic information, historical answer and third-party multi-source data of a user in real time, cleaning and grading abnormal data, and then transmitting the treated multi-source data to S3; S3, carrying out integration analysis based on the multi-source data input in the S2, dynamically adjusting the user portrait weight, generating a standardized user real-time portrait feature vector, and transmitting the user real-time portrait feature vector to the S4; S4, calculating the matching degree of the user portrait and the questionnaire logic rules based on the questionnaire logic rule base synchronized by S1 and the user real-time portrait feature vector transmitted by S3, and outputting a personalized decision result; and S5, receiving the personalized decision result output by the S4, and adapting the user equipment and the age characteristic display questionnaire after calculating the page adaptation effect.
  2. 2. The personalized questionnaire display method based on AI real-time analysis according to claim 1, wherein in S1, the lightweight model is a BERT-Tiny model, and is constructed based on a teacher-student distillation technology, specifically comprising: The teacher model adopts a pre-training original BERT with 12 layers of transformers and 768 dimensions of hidden layers, the student model is simplified into 4/6 layers of transformers, 312 dimensions of hidden layers and a linear projection layer, and 312 dimensions of features are compressed into 128 dimensions of word vectors; The semantic knowledge is transferred through three-layer loss alignment of an embedding layer, a transducer layer and a prediction layer, wherein a mean square error is adopted by an embedding layer distillation loss function, a cross entropy is adopted by a prediction layer distillation loss function, a relation between the embedding layer and the prediction layer is established through fixed mapping, and the two-stage training is carried out through general corpus pre-training and task data fine tuning.
  3. 3. The personalized questionnaire display method based on real-time AI analysis according to claim 2, wherein in S1, the semantic matching degree calculation formula is: ; In the formula, For semantic logic requirements Question label The comprehensive matching degree between the two; Semantic logic requirements entered for a user; a core tag set for a questionnaire topic; For semantic logic requirements 128-Dimensional lightweight word vectors output by the BERT-Tiny model; Is a title label 128-Dimensional lightweight word vectors output by the BERT-Tiny model; the score is precisely matched for the keywords; As the weight coefficient of the light-emitting diode, ; Cosine similarity of two lightweight word vectors; for semantic requirements Question label Is a keyword exact match score.
  4. 4. The personalized questionnaire display method based on real-time AI analysis according to claim 3, wherein in S2, the abnormal data classification processing formula is: ; In the formula, The number of abnormal fields is the number of fields which have errors, deletions and are not in accordance with logic in the acquired data, and the total number of fields is the total number of all fields contained in the acquired data; Repairable anomalies with the anomaly field ratio of <30% adopt neighborhood filling, and invalid anomalies with the anomaly field ratio of more than or equal to 30% adopt independent storage.
  5. 5. The personalized questionnaire display method based on AI real-time analysis as claimed in claim 4, wherein in S3, the user portrait weight is dynamically adjusted, and the formula is: ; In the formula, Is an initial weight; Contributing an increment to the behavior; Is a time decay coefficient; is the first The sub-dimension of each user picture is in Real-time weight of time; the module length of the image feature vector after normalization processing is 1.
  6. 6. The personalized questionnaire display method based on AI real-time analysis according to claim 5, wherein in S4, the matching degree calculation formula of the user portrait and the questionnaire logic rule is: ; In the formula, Is the first The feature vector of the bar rule accords with rule 1 and does not accord with rule 0; Scoring the user feature vector; Is the dimension weight; is the first Feature vector of bar rule Middle (f) The value of the sub-dimension.
  7. 7. The personalized questionnaire display method based on real-time AI analysis according to claim 6, wherein in S5, the evaluation formula of the page adaptation effect is: ; In the formula, Adapting the score for the layout; loading a speed standardized value for the page; The scores are adapted for fonts.
  8. 8. A computer device comprising a processor for coupling with a memory, reading and executing instructions and/or program code in the memory to perform the method of any of claims 1-7.
  9. 9. A computer readable medium, characterized in that the computer readable medium stores computer program code which, when run on a computer, causes the computer to perform the method according to any of claims 1-7.

Description

Questionnaire personalized display method based on AI real-time analysis Technical Field The invention belongs to the technical field of crossing of big data, artificial intelligence and market investigation, and particularly relates to a questionnaire personalized display method based on AI real-time analysis. Background Traditional questionnaires rely on predefined jump rules and question configuration, have complex logic, difficult configuration and insufficient flexibility when the question amount is large, and cannot dynamically adjust investigation contents in combination with real-time situations of users. However, the prior art has the defects that specific question jump rules are set one by one, the configuration difficulty is high in a complex scene, the jump is selected only based on the current questions of a user, real-time data such as geographic positions, equipment information, answering time and the like cannot be integrated, jump paths are optimized only, question expressions and options cannot be dynamically adjusted, and personalized investigation requirements cannot be met. Thus, a new method is needed. Disclosure of Invention The invention aims to provide a questionnaire personalized display method based on AI real-time analysis, which greatly reduces the complexity of questionnaire configuration, improves flexibility, accurately adapts to a user real-time scene, deepens the individuation degree of the questionnaire and enables investigation to be more fit with the actual situation of an answer by semantic rule setting, multi-source data real-time portrait construction and dynamic decision engine optimization. In order to achieve the above purpose, the invention provides a questionnaire personalized display method based on AI real-time analysis, which comprises the following steps: S1, editing a basic questionnaire title through a questionnaire setting module, setting logic requirements by adopting semantic matching degree, calling BERT-Tiny model to analyze the requirements and generating adaptation rules, and constructing a questionnaire logic rule base; s2, collecting answer behaviors, basic information, historical answer and third-party multi-source data of a user in real time, cleaning and grading abnormal data, and then transmitting the treated multi-source data to S3; S3, carrying out integration analysis based on the multi-source data input in the S2, dynamically adjusting the user portrait weight, generating a standardized user real-time portrait feature vector, and transmitting the user real-time portrait feature vector to the S4; S4, calculating the matching degree of the user portrait and the questionnaire logic rules based on the questionnaire logic rule base synchronized by S1 and the user real-time portrait feature vector transmitted by S3, and outputting a personalized decision result; and S5, receiving the personalized decision result output by the S4, and adapting the user equipment and the age characteristic display questionnaire after calculating the page adaptation effect. Preferably, in S1, the BERT-Tiny model is constructed based on a teacher-student distillation technology, and specifically comprises the following steps: The teacher model adopts a pre-training original BERT with 12 layers of transformers and 768 dimensions of hidden layers, the student model is simplified into 4/6 layers of transformers, 312 dimensions of hidden layers and a linear projection layer, and 312 dimensions of features are compressed into 128 dimensions of word vectors; The semantic knowledge is transferred through three-layer loss alignment of an embedding layer, a transducer layer and a prediction layer, wherein a mean square error is adopted by an embedding layer distillation loss function, a cross entropy is adopted by a prediction layer distillation loss function, a relation between the embedding layer and the prediction layer is established through fixed mapping, and the two-stage training is carried out through general corpus pre-training and task data fine tuning. Preferably, in S1, the semantic matching degree calculation formula is: ; In the formula, For semantic logic requirementsQuestion labelThe comprehensive matching degree between the two; Semantic logic requirements entered for a user; a core tag set for a questionnaire topic; For semantic logic requirements 128-Dimensional lightweight word vectors output by the BERT-Tiny model; Is a title label 128-Dimensional lightweight word vectors output by the BERT-Tiny model; the score is precisely matched for the keywords; As the weight coefficient of the light-emitting diode, ;Cosine similarity of two lightweight word vectors; for semantic requirements Question labelIs a keyword exact match score. Preferably, in S2, the abnormal data classification processing formula is: ; In the formula, The number of abnormal fields is the number of fields which have errors, deletions and are not in accordance with logic in the acquired d