User behavior prediction and recommendation system based on big data analysis
By constructing behavior-related graphs and prediction models, the problem of user behavior prediction models relying on data and changing capture is solved, accurate personalized recommendations are achieved, and user satisfaction is improved.
Patent Information
- Application Number
- CN202510391166.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-31
- Publication Date
- 2025-07-18
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing user behavior prediction models require a large amount of high-quality data training, and are difficult to capture user behavior changes and nonlinear relationships, resulting in inaccurate and unfair prediction results.
User data is collected through the data collection module, the data processing module extracts behavior keywords and constructs behavior-related graphs, predicts construction modules for sequence conversion and prediction, and recommends construction modules for encoding replacement and decomposition, and obtains the final selected recommendation vector.
It has achieved a deeper understanding of user interests and preferences, provided accurate and personalized recommendations, and improved user satisfaction and experience.
Smart Images
Figure CN120337916A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of big data analysis, and specifically to a user behavior prediction and recommendation system based on big data analysis. Background Art
[0002] Big data analysis refers to extracting valuable information, insights, and trends from a large amount of data in order to better understand phenomena, support decision-making, and improve business processes. Big data analysis is not just about simple data processing and display, but involves multiple steps and methods, including data collection, preprocessing, data mining, visualization, and model building, etc. Its advantages include better understanding of customer needs, improving operational efficiency, reducing costs, supporting innovation, etc.
[0003] User behavior prediction is a process of modeling and predicting user behavior using technologies such as big data analysis and machine learning. By analyzing users' historical behavior data and personal characteristics, it is possible to predict users' future behavior trends and interest preferences, so as to provide enterprises with more accurate personalized recommendations, customized services, and precision marketing, etc. User behavior prediction has achieved good results in many scenarios, but it also has some disadvantages: the prediction model requires a large amount of high-quality data for training and verification, but the data may be insufficient or incomplete, which may lead to inaccurate prediction results; there are biases in the training data, resulting in unfair or inaccurate prediction results; users' behaviors often change, and the prediction model may be difficult to capture these changes; there may be a non-linear relationship between user behavior and prediction results, which makes it difficult for the prediction model to capture this relationship; therefore, it is of great theoretical and practical significance to continuously improve the model and technology to overcome the disadvantages existing in the process of user behavior prediction and recommendation.
[0004] How to use big data analysis technology to process the collected user data, obtain behavior keywords, and construct a behavior correlation graph, and through the prediction construction of the obtained behavior keywords, obtain a predicted behavior sequence; through the recommendation construction of the obtained behavior keywords and behavior correlation graph, obtain a final selected recommendation vector, is the problem we need to solve; for this reason, a user behavior prediction and recommendation system based on big data analysis is provided now. Summary of the Invention
[0005] The purpose of the present invention can be achieved through the following technical solutions: A user behavior prediction and recommendation system based on big data analysis, including a control center, and the control center is connected with a data collection module, a data processing module, a prediction construction module, and a recommendation construction module; The process of the data collection module collecting user data includes: Set a collection port, which is connected to a number of capture points, and collect the data uploaded by users through the collection port; Capture the collected data at the capture points to obtain user data; the user data includes behavior data, behavior time, attribute data, and feedback data; Associate the behavior time with the corresponding behavior data.
[0006] The process of the data processing module processing user data includes: Extract features from the behavior data in the obtained user data to obtain behavior keywords, and associate the obtained behavior keywords with the behavior time corresponding to the behavior data; Split the feedback data in the obtained user data to obtain a feedback score; Construct a behavior correlation graph based on the obtained behavior keywords, attribute data, and feedback score of the user.
[0007] The process of constructing a behavior correlation graph includes: Set the user as the center point, represent the behavior keywords, attribute data, and feedback score with line segment axes respectively, and mark the obtained several line segment axes; Set unit scales on the line segment axes corresponding to the behavior keywords, and mark the behavior keywords at the unit scales; Obtain a behavior support degree according to the obtained behavior keywords, feedback score, and attribute data.
[0008] The process of predicting and constructing according to the obtained behavior keywords includes: Perform sequence transformation on the obtained behavior keywords to obtain a keyword sequence, sort the obtained keyword sequence according to the associated behavior time, and construct an input sequence set according to the obtained sorted keyword sequence; Perform combined classification on the keyword sequences in the input sequence set to obtain keyword categories, and construct a coding dictionary according to the obtained keyword categories; Search and compare the obtained keyword sequence with the constructed coding dictionary to obtain a unique vector; Search and compare the remaining keyword sequences with the constructed coding dictionary respectively to obtain unique vectors, connect the obtained unique vectors according to the behavior time to obtain a two-dimensional unique coding matrix, and set the obtained two-dimensional unique coding matrix as the input matrix sequence.
[0009] The process of obtaining a predicted behavior sequence includes: Set loop termination conditions and output parameters according to the obtained input matrix sequence, perform hidden processing on the obtained unique vector to obtain an initial vector; Set up torch units and obtain an output sequence through the torch units; Set an activation factor according to the obtained output sequence, and add the single-row vector corresponding to the previous row's time to the output sequence through the activation factor to obtain a second hidden sequence; Repeat the process of obtaining the second hidden sequence according to the obtained output parameters and the loop termination condition, and use the second hidden sequence obtained when the loop termination condition is satisfied as the end hidden sequence at the time of the last row, and mark it as the predicted behavior sequence.
[0010] The process of recommendation construction based on the obtained behavior keywords and behavior-related graph includes: Encode and replace the obtained behavior keywords to obtain keyword encoding vectors, and combine the obtained keyword encoding vectors according to the behavior time to obtain a behavior matrix; Set up a decomposition matrix, and decompose the behavior matrix into a user vector and a behavior item vector through the decomposition matrix; Obtain a behavior interest score based on the behavior support degree and the behavior item vector in the obtained behavior-related graph.
[0011] The process of obtaining the final selected recommendation vector includes: Sort the obtained behavior interest scores to obtain a first-priority behavior vector; Set up an initial behavior matrix according to the obtained first-priority behavior vector, and set parameter factors and a loop stop condition according to the obtained initial behavior matrix; Decompose the obtained initial behavior matrix according to the obtained parameter factors to obtain a user vector and a behavior factor vector; Obtain a behavior interest score based on the obtained behavior support degree and the behavior factor vector, sort the obtained behavior interest scores to obtain a second-priority behavior vector; Set up an initial behavior matrix according to the obtained second-priority behavior vector, and repeat the process of obtaining the second-priority behavior vector until the obtained second-priority behavior vector satisfies the loop stop condition; Mark the obtained second-priority behavior vector that satisfies the loop stop condition as the final selected recommendation vector.
[0012] Compared with the prior art, the beneficial effects of the present invention are: by processing the collected user data, behavior keywords and feedback scores are obtained, and a behavior-related graph is constructed according to the obtained behavior keywords, attribute data and feedback scores; Perform sequence transformation on the obtained behavioral keywords to obtain a keyword sequence, sort the keyword sequence according to the behavioral time to obtain an input sequence set, perform combined classification on the keyword sequences in the input sequence set to obtain keyword categories, construct a coding dictionary based on the obtained keyword categories, perform search and comparison on the keyword sequence and the coding dictionary respectively to obtain a solitary vector, obtain an input matrix sequence according to the behavioral time of the solitary vector, obtain an output sequence through a torch unit, and obtain a second hidden sequence through the set activation factor; until the loop termination condition is satisfied, obtain a predicted behavior sequence; Perform coding replacement on the obtained behavioral keywords to obtain a keyword coding vector, set a decomposition matrix to obtain a behavioral item vector, obtain a behavioral interest score according to the behavioral support degree and the behavioral item vector, sort the obtained behavioral interest scores to obtain a first priority behavior vector, and set an initial behavior matrix and a parameter factor to obtain a behavioral factor vector, obtain a second priority behavior vector according to the behavioral interest score of the behavioral factor vector, until the loop stop condition is satisfied, obtain a final selected recommendation vector; Achieve a deeper understanding of the user's interests and preferences, predict their future behaviors, obtain more valuable and interesting content through accurate personalized recommendations, and improve user satisfaction and experience. Brief Description of the Drawings
[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.
[0014] Figure 1 It is the schematic diagram of the present invention. Detailed Embodiments
[0015] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.
[0016] As Figure 1 shown, a user behavior prediction and recommendation system based on big data analysis includes a control center, and the control center is connected with a data collection module, a data processing module, a prediction construction module, and a recommendation construction module; The data collection module is used to collect user data, and the specific steps include: Set a collection port, which is connected to a number of capture points, and collect the data uploaded by users through the collection port; Mark the uploaded users as i, where i = 1, 2, 3, ……, v1, and v1 is a positive integer; Capture the uploaded data at the capture points to obtain user data, where the user data includes behavior data, behavior time, attribute data, and feedback data; Correspond the obtained behavior data with the behavior time, and associate the behavior time with the corresponding behavior data; It should be further noted that in the specific implementation process, the behavior data includes browsing behavior, purchase behavior, search behavior, comment behavior, and viewing entertainment behavior; the attribute data includes user name, age, gender, geographical location, and occupation; the feedback data includes feedback on prediction results and recommendation results, ratings, and comments; The data processing module is used to process the obtained user data to obtain behavior keywords and construct a behavior correlation graph; Extract features from the behavior data in the obtained user data to obtain behavior keywords, and mark the obtained behavior keywords as XG i ; It should be further noted that in the specific implementation process, one behavior data includes at least one behavior keyword, and the obtained behavior keywords are associated with the behavior time corresponding to the behavior data; Split the feedback data in the obtained user data to obtain a feedback score. It should be further noted that in the specific implementation process, users conduct behavior ratings based on prediction results, speculation results, and comments. The behavior rating is out of 100 points, and individual ratings are given to the prediction result, speculation result, and comment respectively. Add the individual ratings corresponding to the obtained prediction result, speculation result, and comment to obtain the feedback score, and mark the obtained feedback score as FP i ; Construct a behavior correlation graph based on the obtained behavior keywords, attribute data, and feedback score of the user; It should be further noted that in the specific implementation process, the process of constructing a behavior correlation graph includes: Set the user as the center point, and mark the user's name and gender at the center point; Represent the occupation in the attribute data as the first line segment axis, and successively represent the age and position distance in the attribute data as line segment axes. Among them, the position distance is with the collection port as the destination, and mark the distance from the user's geographical location to the destination as the position distance, and mark the obtained position distance as WJ i ; The obtained feedback scores are represented as a line segment axis after the position distance, and the obtained behavior keywords are successively represented as line segment axes; Mark the obtained several line segment axes, and the angles between adjacent two line segment axes are equal; Mark the first line segment axis as j = 1, and then the subsequent line segment axes are successively marked as j = 2, j = 3, ……, v2, where v2 is a positive integer; Mark line segment scales on the line segment axes corresponding to the feedback scores, ages, and position distances respectively; According to the proportional scales corresponding to the obtained feedback scores, ages, and position distances, mark the feedback scores, ages, and position distances at the corresponding line segment scales respectively. For the remaining behavior keywords and the line segment axes corresponding to occupations, taking the unit scale as the standard, mark the behavior keywords and occupations at the unit scales on the corresponding line segment axes; Mark the obtained unit scale as g, where the unit scale is set according to the number of behavior keywords, and mark the number of the obtained behavior keywords as the number of behaviors, denoted as XS i ; It should be further noted that in the specific implementation process, if there are repeated behavior keywords of the same user after collecting user data multiple times, convert the number of repeated behavior keywords into the corresponding number of unit scales, and increase the position of the corresponding repeated behavior keywords on the line segment axis from the original unit scale to the unit scale of the repeated number; for example, if the number of repeated behavior keywords is 5 times, the corresponding unit scale position changes from the original g to 5g; Obtain the behavior support degree according to the obtained behavior correlation diagram; Mark the obtained behavior support degree as XZ i , where XZ i = * *(g*a1 + a1* + * + *g*b1 + ), where a1 represents age, b1 represents the number of repetitions of the behavior keyword adjacent to the feedback score, respectively represent the number of repetitions of the remaining adjacent two behavior keywords.
[0017] The prediction construction module is used to construct a prediction model according to the obtained behavior keywords and obtain a predicted behavior sequence. The specific process includes: According to the obtained behavior keywords, perform sequence transformation on the obtained behavior keywords to obtain a keyword sequence. Sort the obtained keyword sequence in chronological order according to the associated behavior time, and construct an input sequence set based on the keyword sequence sorted in chronological order. Upload the obtained keyword sequence to the input sequence set; According to the keyword sequence in the obtained input sequence set, perform combined classification on the obtained keyword sequence to obtain keyword categories. It should be further noted that in the specific implementation process, the keyword categories include browsing, purchasing, searching, etc.; Construct a coding dictionary according to the obtained keyword categories. It should be further noted that in the specific implementation process, the coding dictionary is composed of several binary code elements. Each keyword category has and only has one coding dictionary, and associate the obtained coding dictionary with the corresponding keyword category; According to the keyword sequence associated with the behavior time, perform a search and comparison between the keyword sequence and the constructed coding dictionary to obtain a solitary vector. In particular, the solitary vector indicates the existence of the behavior corresponding to the keyword sequence, so it is a vector with only one non-zero element; Perform search and comparison between the remaining keyword sequences and the constructed coding dictionary respectively to obtain solitary vectors. Connect the obtained solitary vectors in chronological order according to the behavior time to obtain a two-dimensional solitary coding matrix of behavior data, and set the obtained two-dimensional solitary coding matrix as the input matrix sequence; Set loop parameters according to the obtained input matrix sequence. The loop parameters include output parameters and loop termination conditions; Perform a hiding process on the obtained solitary vector to obtain an initial vector. It should be further noted that in the specific implementation process, the hiding process is to multiply an arbitrarily selected solitary vector by 0 to obtain a zero vector, and mark the obtained zero vector as the initial vector; Set a torch unit, which is used to multiply the solitary vector at the current behavior time by the two-dimensional solitary coding matrix to obtain an output sequence; Set an activation factor according to the obtained output sequence, and add the solitary vector corresponding to the previous behavior time to the output sequence through the activation factor to obtain a second hidden sequence; Repeat the process of obtaining the second hidden sequence according to the obtained output parameters and loop termination conditions. When the loop termination condition is met, use the obtained second hidden sequence as the end hidden sequence at the last behavior time; According to the obtained end hidden sequence, mark the obtained end hidden sequence as the predicted behavior sequence for the user's next behavior time; It should be further noted that, in the specific implementation process, the obtained end hidden sequence is obtained in sequence according to the chronological order of the behavior time of the solo vector, which can capture the chronological order and relevance between user behaviors, and more accurately predict the predicted behavior sequence of the user's next behavior time.
[0018] The recommended construction module is used to construct a recommendation model according to the obtained behavior keywords and behavior-related graphs, and obtain a final selected recommendation vector according to the constructed recommendation model. The specific process includes: According to the obtained behavior keywords, encode and replace the obtained behavior keywords to obtain keyword encoding vectors, and combine the obtained keyword encoding vectors in the chronological order of behavior time to obtain a behavior matrix; It should be further noted that, in the specific implementation process, the encoding replacement is to convert the obtained behavior keywords into vectors containing binary code element sequences; Set a decomposition matrix, and decompose the obtained behavior matrix into a user vector and a behavior item vector through the decomposition matrix, and mark the obtained behavior item vector as d; Obtain a behavior interest score according to the behavior support degree and behavior item vector in the obtained behavior-related graph, and mark the obtained behavior interest score as XT i , where XT i =XZ i *d'+(α1*a1* +α2* * )* * , d' represents the modulus length of the behavior item vector d, α1 and α2 are proportionality factors, and α1 + α2 = 1; Sort the obtained behavior interest scores in descending order, and mark the behavior item vectors ranked in the top k1 as the first priority behavior vectors; Set an initial behavior matrix according to the obtained first priority behavior vectors, and set parameter factors according to the obtained initial behavior matrix; Decompose the obtained initial behavior matrix according to the obtained parameter factors to obtain a user vector and a behavior factor vector; Obtain a behavior interest score according to the behavior support degree and behavior factor vector in the obtained behavior-related graph, sort the obtained behavior interest scores in descending order, and mark the behavior factor vectors ranked in the top k2 as the second priority behavior vectors, and k1 > k2; Then set an initial behavior matrix according to the obtained second priority behavior vectors, and repeat the process of obtaining the second priority behavior vectors until the obtained second priority behavior vectors meet the loop stop condition. The loop stop condition is k pThe number satisfies the set loop stop condition, where p = 1, 2, 3, ……, v3, and v3 is a positive integer; Mark the second priority behavior vector that satisfies the loop stop condition as the final selected recommendation vector, and recommend the obtained final selected recommendation vector to the user.
[0019] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation manners. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.
Claims
1. A user behavior prediction and recommendation system based on big data analysis, including a control center, characterized in that, The control center is connected to a data collection module, a data processing module, a prediction construction module, and a recommendation construction module; The data collection module is used to collect user data; The data processing module is used to process the obtained user data, obtain behavior keywords, and construct a behavior correlation graph; The prediction construction module is used to perform prediction construction based on the obtained behavior keywords to obtain a predicted behavior sequence; The recommendation construction module is used to perform recommendation construction based on the obtained behavior keywords and behavior correlation graph to obtain a final selected recommendation vector.
2. The user behavior prediction and recommendation system based on big data analysis according to claim 1, wherein The user data includes behavior data, behavior time, attribute data, and feedback data. The process of the data collection module collecting user data includes: Setting a collection port. The collection port is connected to a number of capture points, and the data uploaded by the user is collected through the collection port; Capturing the collected data at the capture points to obtain user data; Associating the behavior time with the corresponding behavior data.
3. The user behavior prediction and recommendation system based on big data analysis according to claim 2, characterized in that, The process of the data processing module processing user data includes: Performing feature extraction on the behavior data in the obtained user data to obtain behavior keywords, and associating the obtained behavior keywords with the behavior time corresponding to the behavior data; Splitting the feedback data in the obtained user data to obtain a feedback score; Constructing a behavior correlation graph based on the obtained behavior keywords, attribute data, and feedback score of the user.
4. The user behavior prediction and recommendation system based on big data analysis according to claim 3, characterized in that, The process of constructing a behavior correlation graph includes: Setting the user as the center point, representing the behavior keywords, attribute data, and feedback score with line segment axes respectively, and marking the obtained several line segment axes; Setting unit scales on the line segment axes corresponding to the behavior keywords, and marking the behavior keywords at the unit scales; Obtaining a behavior support degree based on the obtained behavior keywords, feedback score, and attribute data.
5. A user behavior prediction and recommendation system based on big data analysis according to claim 4, characterized in that, The process of performing prediction construction based on the obtained behavior keywords includes: Performing sequence transformation on the obtained behavior keywords to obtain a keyword sequence, sorting the obtained keyword sequence according to the associated behavior time, and constructing an input sequence set based on the obtained sorted keyword sequence; Performing combined classification on the keyword sequences in the input sequence set to obtain keyword categories, and constructing a coding dictionary based on the obtained keyword categories; Searching and comparing the obtained keyword sequence with the constructed coding dictionary to obtain a unique vector; Searching and comparing the remaining keyword sequences with the constructed coding dictionary respectively to obtain unique vectors, connecting the obtained unique vectors according to the behavior time to obtain a two-dimensional unique coding matrix, and setting the obtained two-dimensional unique coding matrix as an input matrix sequence.
6. The user behavior prediction and recommendation system based on big data analysis according to claim 5, characterized in that, The process of obtaining a predicted behavior sequence includes: Setting a loop termination condition and output parameters according to the obtained input matrix sequence, performing hidden processing on the obtained unique vector to obtain an initial vector; Setting a torch unit, and obtaining an output sequence through the torch unit; Setting an activation factor according to the obtained output sequence, and adding the unique vector corresponding to the previous behavior time to the output sequence through the activation factor to obtain a second hidden sequence; Repeat the process of obtaining the second hidden sequence according to the obtained output parameters and the loop termination condition. Use the second hidden sequence obtained when the loop termination condition is met as the end hidden sequence of the last behavior time, and mark it as the predicted behavior sequence.
7. A user behavior prediction and recommendation system based on big data analysis according to claim 6, characterized in that, The process of recommendation construction based on the obtained behavior keywords and behavior-related graphs includes: Encode and replace the obtained behavior keywords to obtain keyword encoding vectors, and combine the obtained keyword encoding vectors according to the behavior time to obtain a behavior matrix. Set a decomposition matrix, and decompose the behavior matrix into a user vector and a behavior item vector through the decomposition matrix. Obtain the behavior interest score based on the behavior support degree and the behavior item vector in the obtained behavior-related graph.
8. A user behavior prediction and recommendation system based on big data analysis according to claim 7, characterized in that, The process of obtaining the final selected recommendation vector includes: Sort the obtained behavior interest scores to obtain the first priority behavior vector. Set an initial behavior matrix according to the obtained first priority behavior vector, and set parameter factors and a loop stop condition according to the obtained initial behavior matrix. Decompose the obtained initial behavior matrix according to the obtained parameter factors to obtain a user vector and a behavior factor vector. Obtain the behavior interest score based on the obtained behavior support degree and the behavior factor vector, sort the obtained behavior interest scores, and obtain the second priority behavior vector. Set an initial behavior matrix according to the obtained second priority behavior vector, and repeat the process of obtaining the second priority behavior vector until the obtained second priority behavior vector meets the loop stop condition. Mark the obtained second priority behavior vector that meets the loop stop condition as the final selected recommendation vector.