Search

CN-122022870-A - AI intelligent marketing scene construction method based on user behavior track big data

CN122022870ACN 122022870 ACN122022870 ACN 122022870ACN-122022870-A

Abstract

The invention discloses an AI intelligent marketing scene construction method based on big data of user behavior track, relating to the technical field of AI intelligent marketing scene construction, comprising the following steps of extracting interaction density residual factors, content semantic focus offset values and interest anchor point residual shadow intensities in track sections which are judged to continuously generate path interruption and do not form complete behavior chains, and forming jump behavior track residual groups to determine behavior track data; mapping the jump behavior trace residual group, constructing a three-dimensional behavior coherence estimation tensor consisting of jump semantic contracture rate, behavior target convergence and anchor point landing potential, and calculating a behavior coherence index parameter. The invention solves the problem that intention is difficult to identify when the path is interrupted and no continuous chain exists in the user behavior track, and realizes continuous intention identification and intelligent marketing triggering of fragmented behavior data.

Inventors

  • YU LINYI

Assignees

  • 上海海湃领客文化科技有限公司

Dates

Publication Date
20260512
Application Date
20260225

Claims (9)

  1. 1. The AI intelligent marketing scene construction method based on the big data of the user behavior track is characterized by comprising the following steps: S1, analyzing a user behavior track section by constructing a jump recognition rule set comprising behavior transfer sparsity, jump target anti-semantic ratio and behavior sequence progressive breakpoint prominence, and judging whether a path interruption continuously occurs in the user behavior track and a complete behavior chain is not formed; s2, extracting interaction density residual factors, content semantic focus offset values and interest anchor point residual image intensities in a track section which is judged to be continuous in occurrence of path interruption and not to form a complete behavior chain, and forming a jump behavior track residual group so as to determine behavior track data; s3, mapping the jump behavior trace residual group, constructing a three-dimensional behavior coherence estimation tensor consisting of jump semantic contracture rate, behavior target convergence and anchor point landing potential, and calculating a behavior coherence index parameter; S4, based on the behavior coherence index parameters, fusing behavior target reproduction degree, behavior orientation convergence stability and potential label aggregation reliability to generate an intention focusing judgment vector so as to judge whether a user has continuous behavior intention which can be used for constructing a marketing scene; S5, classifying the behavior track into three types of high confidence intentions, intentions to be confirmed and unidentifiable intentions according to the judging result, and executing corresponding behavior trigger processing operation based on the classifying result.
  2. 2. The AI-intelligent marketing scenario construction method based on the user behavior trace data of claim 1, wherein S1 specifically comprises the steps of: S101, acquiring a continuous behavior node sequence in a user behavior track, and aiming at each pair of adjacent behavior nodes, calculating behavior transfer sparsity, jump target anticreep ratio and behavior sequence progressive breakpoint prominence, wherein the behavior transfer sparsity is determined according to rareness of behavior types in a global track, the jump target anticreep ratio is calculated according to semantic vector included angle values of adjacent behavior node target contents, and the behavior sequence progressive breakpoint prominence is generated by joint modeling according to behavior time interval difference values and behavior content change rates and is used for constructing a jump recognition rule set; S102, respectively applying behavior transfer sparsity, jump target anticreep ratio and behavior sequence progressive breakpoint prominence in jump recognition rule set to each behavior node pair of a user behavior track, and marking the behavior node pair as a path interruption behavior node pair if values of any two of the behavior transfer sparsity, the jump target anticreep ratio and the behavior sequence progressive breakpoint prominence are simultaneously in a preset jump recognition threshold interval; s103, when a continuous path interruption behavior node pair appears in the user behavior track and a semantic progressive relationship or a time adjacent relationship does not exist between any behavior nodes in the continuous segment, judging that the condition that the path interruption continuously occurs in the user behavior track and a complete behavior chain is not formed is established.
  3. 3. The AI-intelligent marketing scenario construction method based on the user behavior trace data of claim 2, wherein S101 specifically comprises: extracting a continuous behavior node sequence from a user behavior track according to a time sequence, representing each behavior node as a node record containing behavior types, behavior target contents and behavior occurrence time, and forming a behavior node pair by using adjacent behavior nodes as a basic data unit for calculating behavior transfer sparsity, jump target anticreep ratio and behavior sequence progressive breakpoint prominence; for each behavior node pair, calculating behavior transfer sparsity based on occurrence distribution of behavior types in a global behavior track, calculating jump target anticreep ratio based on semantic vector included angle numerical values of target contents of adjacent behavior nodes, and generating behavior sequence progressive breakpoint prominence based on joint modeling of behavior time interval difference values corresponding to the adjacent behavior nodes and behavior content change rates; And combining the corresponding behavior transfer sparsity, the jump target anti-semantic ratio and the behavior sequence progressive breakpoint prominence of the same behavior node pair according to a unified data mapping rule to form a jump recognition rule item aiming at the behavior node pair, and assembling a plurality of jump recognition rule items to construct a jump recognition rule set for subsequent section-by-section analysis of the user behavior track.
  4. 4. The AI-intelligent marketing scenario construction method based on the user behavior trace data of claim 1, wherein S2 specifically comprises the steps of: s201, based on the user behavior track which is judged to continuously generate path interruption and does not form a complete behavior chain, carrying out time sequence aggregation on path interruption behavior node pairs, extracting track fragments formed by the continuous path interruption behavior node pairs, and determining the track fragments as track fragments which continuously generate the path interruption and do not form the complete behavior chain, wherein the track fragments are used as an analysis range for extracting subsequent factors; S202, in a track segment, extracting interaction density residual factors based on interaction frequency and stay time distribution of each behavior node, extracting content semantic focus offset values based on offset amplitude of semantic barycenter of behavior target content in the track segment relative to semantic barycenter of an effective behavior chain before the track segment, and extracting interest anchor point residual image intensity based on semantic similarity distribution between the behavior nodes in the track segment and a user historical interest anchor point; S203, combining the interaction density residual factor, the content semantic focus offset value and the interest anchor point residual image intensity according to the unified data dimension to form a jump behavior track residual group corresponding to the track section, and taking the jump behavior track residual group as an input basis for determining the behavior track data of the track section.
  5. 5. The AI-intelligent marketing scenario construction method based on the user behavior trace data of claim 4, wherein S202 is specifically: in the track section, the interaction frequency and the residence time length corresponding to each behavior node are obtained, the interaction frequency and the residence time length of each behavior node in the track section are distributed and counted, the interaction frequency and the residence time length are compared with the historical interaction distribution of a user in a non-track section, the residual change characteristics of the interaction distribution in the track section relative to the historical interaction distribution are extracted, and an interaction density residual factor is formed; Calculating semantic centers of behavior target contents in the track section based on target content semantic vectors of all behavior nodes in the track section, calculating corresponding semantic centers based on the target content semantic vectors of the behavior nodes in the effective behavior chain in front of the track section, and extracting content semantic focus offset values by comparing offset amplitudes between the two semantic centers; and calculating similarity distribution between the target content semantic vector of each behavior node and the interest anchor semantic vector in the track section based on the interest anchor set formed in the user history behavior, and extracting the strength of the interest anchor ghost according to the concentration degree and the attenuation characteristic of the similarity distribution.
  6. 6. The AI-intelligent marketing scenario construction method based on the user behavior trace big data of claim 1, wherein S3 is specifically: Mapping the interaction density residual factors, the content semantic focus offset values and the interest anchor point residual image intensities contained in the jump behavior track residual group, and mapping the three factors to a behavior semantic space through a unified feature embedding rule to form semantic feature representation for consistency speculation; Calculating jump semantic contracture rate based on semantic feature representation after mapping processing, wherein the jump semantic contracture rate is obtained by comparing the contraction degree of a semantic feature distribution range in a track segment relative to an effective behavior chain semantic distribution range in front of the track segment, calculating behavior target convergence degree which is obtained by counting the aggregation degree of behavior target semantic vectors in the track segment in a semantic space, calculating anchor point potential which is obtained by approaching trend of semantic features in the track segment to the user historical interest anchor point semantic center, and combining the jump semantic contracture rate, the behavior target convergence degree and the anchor point potential according to a fixed dimension sequence; and constructing a three-dimensional behavior coherent presumption tensor based on the combined jump semantic contracture rate, the behavior target convergence degree and the anchor point potential, and carrying out weighted fusion calculation on the three-dimensional behavior coherent presumption tensor to generate a behavior coherent index parameter corresponding to the track segment.
  7. 7. The AI-intelligent marketing scenario construction method based on the user behavior trace data of claim 1, wherein S4 specifically comprises the steps of: S401, based on the behavior coherence index parameters, counting the frequency and semantic similarity of each behavior target in the track section in the user history behavior track, and generating the behavior target recurrence degree; based on the semantic migration path of the continuous behavior targets in the behavior sequence, evaluating the stable convergence interval of the behavior directions in the semantic space to generate behavior direction convergence stability; S402, carrying out dimension unified processing on behavior target reproduction degree, behavior orientation convergence stability, potential tag aggregation reliability and behavior coherence index parameters, and carrying out embedded mapping according to a preset combination sequence to generate an intention focusing discrimination vector for describing the intention focusing degree of the user behavior; S403, comparing the intention focusing discrimination vector with a preset continuous behavior intention discrimination interval, and judging that a user has continuous behavior intention capable of being used for constructing a marketing scene when the value of the intention focusing discrimination vector in each dimension simultaneously falls into a corresponding discrimination interval range.
  8. 8. The AI-intelligent marketing scenario construction method based on the user behavior trace data of claim 7, wherein S401 is specifically: Determining an analysis track section based on the behavior coherence index parameters, counting the occurrence frequency of each behavior target in the track section in a user historical behavior track, synchronously calculating the similarity distribution between the semantic vectors of the behavior targets in the track section and the semantic vectors of the historical behavior targets, and generating the behavior target recurrence degree of the corresponding track section by carrying out joint normalization on the occurrence frequency and the semantic similarity; Based on a time sequence formed by continuous behavior targets in a track segment, extracting a migration path of a semantic vector of an adjacent behavior target in a behavior semantic space, carrying out statistical analysis on a direction change amplitude and a concentrated interval of the migration path, determining a stable convergence interval of behavior pointing in the semantic space, and generating behavior pointing convergence stability; And calculating the aggregation density of semantic matching distribution in the potential label set in the track segment based on the matching relation between the semantic vector of the behavioral target in the track segment and each label semantic vector in the user history potential label set, and generating the potential label aggregation credibility by combining the concentration degree of the aggregation distribution.
  9. 9. The AI-intelligent marketing scenario construction method based on the user behavior trace data of claim 1, wherein S5 is specifically: According to the numerical expression of the intention focusing discrimination vector on each dimension, comparing with a preset continuous behavior intention discrimination interval, generating an intention discrimination result parameter corresponding to the behavior track according to the position relation of each dimension in the interval; Performing a first class of behavior trigger processing operation corresponding to the behavior track classified as the high confidence intention in the marketing scene, wherein the first class of behavior trigger processing operation comprises the steps of activating a precise pushing module, loading high-matching content resources and connecting an interest recall strategy in parallel; Based on user behavior feedback data after execution of various behavior trigger processing operations, updating a behavior track database and supplementing an intention recognition training sample, and improving adaptability and accuracy of behavior track classification and behavior trigger processing operations.

Description

AI intelligent marketing scene construction method based on user behavior track big data Technical Field The invention relates to the technical field of AI intelligent marketing scene construction, in particular to an AI intelligent marketing scene construction method based on big data of user behavior tracks. Background The AI intelligent marketing scene construction based on the user behavior track big data refers to that through collecting and analyzing multidimensional behavior track data (such as click paths, browsing time length, page jumps, geographic positions, equipment information, interaction frequency and the like) of users on a digital platform, the user intention and behavior patterns are deeply mined by means of big data processing and artificial intelligent technology, so that personalized and dynamic marketing scenes are automatically constructed, and accurate touch and intelligent pushing are realized. In the prior art, the user behaviors are recorded in real time mainly through a data acquisition system, raw behavior data are cleaned, aggregated and feature extracted by utilizing a big data platform, then a machine learning model (such as a classification model, a clustering model, a sequence prediction model and the like) is combined to generate user portraits and behavior predictions, marketing contents and a trigger mechanism are matched according to prediction results, and a multi-scene marketing logic chain centering on user intention is constructed. The method comprises the following core links of firstly collecting and preprocessing behavior data to ensure the integrity and effectiveness of the data, secondly constructing user figures to form a label system and continuously updating, thirdly training and reasoning AI models to identify the current conversion stage or interest trend of the user, fourthly generating and dispatching a marketing scene, intelligently combining marketing content and a touch mode according to the identification result, and fifthly, feeding back marketing effects and self-optimizing the models to realize closed-loop learning and strategy iteration. The key of the technology is that a static label system is evolved into an intelligent marketing network driven by dynamic behaviors, so that the accuracy and instantaneity of marketing response are greatly improved. The prior art has the following defects: in the process of realizing the AI intelligent marketing scene construction based on the large data of the user behavior track, when the user uses the system in a fragmentation mode, the current behavior path is frequently interrupted and is transferred to other behavior paths which are not related to each other, and the behavior track with continuity is not formed, the behavior data of the user is expressed as a discrete state which lacks stable context association. In this case, since there is no explicit behavior acceptance relationship and intent convergence feature in the user behavior trace, it is difficult for the system to establish efficient interest delivery and intent evolution decision logic based on the behavior trace. The conventional AI intelligent marketing scene construction technology based on the big data of the user behavior track can not accurately judge whether the user has continuous behavior intention which can be used for constructing the marketing scene according to the behavior track data of the situation that the user continuously generates path interruption in the behavior track and does not form a complete behavior chain, so that the user behavior is erroneously identified as invalid or abnormal behavior, and cannot participate in subsequent marketing scene construction and triggering, and further adverse effects such as reduced accuracy of user intention identification, insufficient coverage of marketing scene construction, reduced overall marketing intelligent effect and the like are caused. The above information disclosed in the background section is only for enhancement of understanding of the background of the disclosure and therefore it may include information that does not form the prior art that is already known to a person of ordinary skill in the art. Disclosure of Invention The invention aims to provide an AI intelligent marketing scene construction method based on big data of user behavior tracks, so as to solve the problems in the background technology. In order to achieve the purpose, the invention provides the following technical scheme that the AI intelligent marketing scene construction method based on the big data of the user behavior track specifically comprises the following steps: S1, analyzing a user behavior track section by constructing a jump recognition rule set comprising behavior transfer sparsity, jump target anti-semantic ratio and behavior sequence progressive breakpoint prominence, and judging whether a path interruption continuously occurs in the user behavior track and a complete behavior chain i