A user portrait construction method, device, equipment and storage medium
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING DEEP DOT INTELLIGENCE TECH CO LTD
- Filing Date
- 2026-04-08
- Publication Date
- 2026-07-03
Smart Images
Figure CN122335342A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, device and storage medium for constructing user profiles. Background Technology
[0002] With the rapid development of digital services, user profiling has become a crucial foundation for operators' precision marketing, personalized services, and business decisions. Its accuracy and timeliness directly impact business performance. High-quality user profiles not only reflect users' basic attributes and consumption habits but also depict their interests, behavioral patterns, and potential needs.
[0003] Currently, existing user profiling methods mostly rely on feature extraction and tag generation based on single-dimensional data, or simply integrate data from different sources, lacking deep fusion of multi-source heterogeneous data. The lack of systematic correlation analysis between different data dimensions results in an inability to comprehensively and accurately depict user behavior characteristics. Therefore, existing technologies still have significant limitations in multi-source data fusion, and the resulting user profiles fail to fully reflect users' true behavioral patterns and dynamic preferences, thus limiting the application value and accuracy of these profiles. Summary of the Invention
[0004] This application provides a user profile construction method, apparatus, device, and storage medium, which enables the constructed user profile to more comprehensively and accurately reflect the user's behavioral characteristics and preferences, thereby improving the accuracy and reliability of the user profile.
[0005] In a first aspect, embodiments of this application provide a user profile construction method, the method comprising: Collect multi-source user behavior data, which includes communication-side data and business application-side data; Feature extraction is performed on the multi-source user behavior data to generate an initial feature dataset; The multi-source features in the initial feature dataset are fused to obtain a fused feature dataset; Based on the fused feature dataset, a set of user tags is generated and a user profile is constructed.
[0006] One feasible implementation method, wherein fusing the multi-source features in the initial feature dataset to obtain a fused feature dataset, includes: The multi-source features in the initial feature dataset are weighted and summed to obtain a comprehensive feature value; The multi-source features in the initial feature dataset are subjected to association rule mining to obtain associated features; The integrated feature values and the associated features are fused together to obtain a fused feature dataset.
[0007] One feasible implementation, wherein the weighted summation of multi-source features in the initial feature dataset to obtain comprehensive features, includes: Based on the data stability and data integrity of each feature in the multi-source features, the credibility weight of each feature in the multi-source features is determined; The observed values of each feature in the multi-source features are standardized to determine the standard feature value of each feature in the multi-source features; Based on the credibility weight of each feature in the multi-source features, the standard feature values of each feature in the multi-source features are weighted and summed to obtain the comprehensive feature value corresponding to the multi-source features.
[0008] One feasible implementation involves performing association rule mining processing on the multi-source features in the initial feature dataset to obtain associated features, including: Based on the multi-source features in the initial feature dataset, a transaction set is constructed, which includes multiple feature items of the user. Based on a preset mining algorithm, multiple feature items in the transaction set are mined to determine the frequent feature item set in the transaction set; The combination of feature terms with a confidence level not less than a preset confidence threshold is extracted from the set of frequent feature terms and determined as associated features.
[0009] One feasible implementation, wherein determining the credibility weight of each feature in the multi-source features based on the data stability and data integrity of each feature, includes: Obtain the observation value and data record of each feature in the multi-source features within the historical time window; Based on the standard deviation of the observed values, the stability score of the data source corresponding to each feature is determined; Based on the data records, the number of historical time windows without missing values is counted, and the ratio of the number of historical time windows without missing values to the total number of historical time windows is determined as the integrity score of the data source corresponding to each feature. The stability score and the integrity score are weighted and summed to determine the credibility weight of each feature in the multi-source features.
[0010] One feasible implementation, wherein extracting feature item combinations with a confidence level not less than a preset confidence threshold from the set of frequent feature items and determining them as associated features, includes: Based on the set of frequent feature terms, candidate feature term combinations are generated; Based on the candidate feature combination, the confidence level of the candidate feature combination is calculated; The feature items with a confidence level not less than a preset confidence threshold are combined and determined as associated features.
[0011] One feasible implementation, wherein generating a user tag set and constructing a user profile based on the fused feature dataset, includes: Based on the fused feature dataset, multiple tags included in the preset tag library are scored to obtain the score value corresponding to each tag; The tags with scores greater than a preset score threshold are filtered to obtain an initial tag set; The score values corresponding to each tag in the initial tag set are updated based on the time decay factor to obtain the updated score values of each tag; The updated rating values of the tags that are greater than the preset rating threshold are filtered to obtain a set of user tags. The user profile is constructed based on the user tag set.
[0012] Secondly, embodiments of this application provide a user profile building apparatus, including: The data acquisition module is used to collect multi-source user behavior data, which includes communication-side data and business application-side data. The feature extraction module is used to extract features from the multi-source user behavior data and generate an initial feature dataset; The feature fusion module is used to fuse the multi-source features in the initial feature dataset to obtain a fused feature dataset; The profile building module is used to generate a set of user tags and build user profiles based on the fused feature dataset.
[0013] Thirdly, embodiments of this application provide an electronic device, the device including: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store a program, which includes instructions that, when executed by the processor, cause the processor to perform any of the implementation steps of the user profile construction method described above.
[0014] Fourthly, embodiments of this application provide a computer-readable storage medium for storing a computer program, which, when executed by a terminal device, implements any of the implementation steps of the user profile construction method described above.
[0015] As can be seen from the above technical solutions, the embodiments of this application have the following advantages: As can be seen from the above technical solution, this application provides a user profile construction method, apparatus, device, and storage medium. The method includes: first, collecting multi-source user behavior data, wherein the multi-source user behavior data includes communication-side data and business application-side data. Then, extracting features from the multi-source user behavior data to generate an initial feature dataset. Next, fusing the multi-source features in the initial feature dataset to obtain a fused feature dataset. Finally, based on the fused feature dataset, generating a user tag set and constructing a user profile.
[0016] As can be seen, this solution collects multi-source user behavior data, including communication-side data and business application-side data, extracts features from these sources to generate an initial feature dataset, and then fuses the multi-source features in the initial feature dataset to obtain a fused feature dataset. Finally, user tags are generated and user profiles are constructed based on this fused feature dataset. Through this multi-source feature fusion approach, this solution effectively integrates data from different sources, overcoming the data fragmentation problem in existing technologies. This allows the constructed user profile to more comprehensively and accurately reflect user behavior characteristics and preferences, thereby improving the accuracy and reliability of the user profile. Attached Figure Description
[0017] Figure 1 A flowchart illustrating a user profile construction method provided in this application embodiment; Figure 2 This is a schematic diagram of a user profile building device provided in an embodiment of this application. Detailed Implementation
[0018] As mentioned earlier, current user profiling methods primarily rely on feature extraction and tag generation from single-dimensional data, or simply integrate data from different sources, lacking deep fusion of multi-source heterogeneous data. The lack of systematic correlation analysis between different data dimensions prevents a comprehensive and accurate depiction of user behavior characteristics. Therefore, existing technologies still have significant limitations in multi-source data fusion, and the resulting user profiles fail to fully reflect users' true behavioral patterns and dynamic preferences, thus limiting the applicability and accuracy of these profiles.
[0019] Meanwhile, in the feature extraction process, existing technologies have failed to fully explore the potential relationships between multi-source data, resulting in low distinguishability of the extracted features, making it difficult to characterize fine-grained differences between users, and thus affecting the accuracy and reliability of user profiles.
[0020] To address the aforementioned issues, this application provides a user profile construction method. First, multi-source user behavior data is collected, including communication-side data and business application-side data. Then, features are extracted from the multi-source user behavior data to generate an initial feature dataset. Next, the multi-source features in the initial feature dataset are fused to obtain a fused feature dataset. Finally, based on the fused feature dataset, a user tag set is generated, and a user profile is constructed.
[0021] As can be seen, this solution collects multi-source user behavior data, including communication-side data and business application-side data, extracts features from these sources to generate an initial feature dataset, and then fuses the multi-source features in the initial feature dataset to obtain a fused feature dataset. Finally, user tags are generated and user profiles are constructed based on this fused feature dataset. Through this multi-source feature fusion approach, this solution effectively integrates data from different sources, overcoming the data fragmentation problem in existing technologies. This allows the constructed user profile to more comprehensively and accurately reflect user behavior characteristics and preferences, thereby improving the accuracy and reliability of the user profile.
[0022] Furthermore, this solution further explores the potential correlations between multi-source data during the fusion process of the initial feature dataset, performing correlation analysis on features from different sources to strengthen the intrinsic connections between multi-source features and improve the discriminative power of feature representation. Based on this, it can effectively characterize fine-grained differences between users, addressing the shortcomings of existing technologies in multi-source data correlation mining, and further improving the accuracy and reliability of user profiles.
[0023] It should be noted that the embodiments of this application do not limit the executing entity of the user profiling method. For example, the user profiling construction method of this application embodiment can be applied to information processing devices such as servers or terminal devices. The server can be a standalone server, a cluster server, or a cloud server. The terminal device can be an electronic device such as a smartphone, computer, personal digital assistant (PDA), or tablet computer.
[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0025] Figure 1This is a flowchart illustrating a user profile construction method provided in an embodiment of this application. (Combined with...) Figure 1 As shown, it may include steps S101-S104.
[0026] S101: Collect multi-source user behavior data, which includes communication-side data and business application-side data.
[0027] This application embodiment collects multi-source user behavior data, which includes communication-side data and business application-side data.
[0028] Specifically, communication-side data refers to user behavior data within the communication network, including base station location information, base station handover records, signaling interaction data, dwell time, location area distribution information, and uplink and downlink traffic data. Service-side data refers to user behavior data in various applications, including APP access data and mini-program access data.
[0029] It should be noted that the embodiments of this application can use tools such as Flume and Kafka (a distributed data collection and message transmission middleware) to collect multi-source user behavior data, such as user communication side data and business application side data, in real time.
[0030] S102: Extract features from multi-source user behavior data to generate an initial feature dataset.
[0031] In this embodiment of the application, after completing the collection of multi-source user behavior data, Spark is first used to clean the multi-source user behavior data, such as deduplication, filling in missing values, and format standardization. After completing the data preprocessing, feature extraction is further performed on the processed multi-source user behavior data to generate an initial feature dataset.
[0032] Specifically, by extracting features from communication-side data and business application-side data, features such as "base station handover count", "percentage of frequently accessed APP types" and "mini-program function usage preference vector" can be obtained, thereby constructing an initial feature dataset.
[0033] S103: Perform fusion processing on the multi-source features in the initial feature dataset to obtain the fused feature dataset.
[0034] This application's embodiments determine the credibility weight of each feature in a multi-source feature set based on the data stability and data integrity of each feature. Specifically, it obtains the observation values of each feature within a historical time window. ,in, Features Corresponding data source These are observations within a historical time window t. Based on the observations... Standard deviation Determine features Corresponding data source Stability score : ; in, i For the first feature in the initial feature dataset i Multiple source features, Features Corresponding data source Observations within historical time windows The standard deviation is used to reflect the data source that generated this feature. stability, m This represents the number of multi-source features in the initial feature dataset used to calculate the standard deviation.
[0035] For example, if , , Then the stability score for: ; like Then the stability score is... for: ; If the standard deviations are all the same, that is Then at this time This indicates that stability has no differential effect on the credibility weights in subsequent calculations.
[0036] In addition, data records for each feature in the multi-source features are obtained within a historical time window. Based on these data records, the number of historical time windows without missing values is counted. The ratio of the number of historical time windows without missing values to the total number of historical time windows is used to obtain the data source integrity score for each feature. .
[0037] Based on this, the stability score and integrity score Perform a weighted summation and assume a stability score. The corresponding first weighting coefficient is 0.5, and the integrity score is... The corresponding second weight coefficient is also 0.5, thus determining the confidence weight of each feature in the multi-source features. for: ; For example, suppose , , , Then the credibility weight It is 0.736. It is 0.264.
[0038] Furthermore, the observed values of each feature in the multi-source features are standardized to determine the standard feature value of each feature. Specifically, this involves standardizing the observed values of each feature within the historical time window. Standardization is performed, mapping the values to the interval [0,1] to eliminate the influence of dimensions and obtain standard eigenvalues. : ; in, The observation values for the current time window. These are observations within a historical time window.
[0039] Based on this, and using credibility weights With standard eigenvalues The weighted sum of the standard feature values of each feature in the multi-source features is used to obtain the comprehensive feature value corresponding to the multi-source features. : ; For example, in It is 0.736. It is 0.8, and It is 0.264. When the value is 0.75, the comprehensive eigenvalue is... for: ; Furthermore, association rule mining is performed on the multi-source features in the initial feature dataset to obtain associated features. Specifically, a transaction set is constructed based on the multi-source features in the initial feature dataset. , K For the set of transactions T The total number of transactions in the middle, and satisfying , It is a set of feature terms, containing multiple positive terms. N This represents the total number of feature terms.
[0040] Based on a pre-defined mining algorithm, multiple feature items in the transaction set are mined to determine the frequent feature item set in the transaction set. Specifically, when the mining algorithm uses the Apriori algorithm, it first traverses all feature items. Calculate the support of each feature item support({ }), and filter out the support ({ }) greater than or equal to the first threshold The feature terms constitute the initial frequent feature term set. Its expression is: ; For example, in the total number of transactions K When the value is 1000, it includes If the number of transactions is 150, then at this time... support({ })for: ; If the first threshold If it is 0.1, then The support level is greater than the first threshold, therefore .
[0041] Based on this, the initial frequent feature set Connect the feature terms in pairs to generate candidate feature term combinations. The connection constraint is that the first m-2 terms are the same. Based on this candidate feature term combination, the candidate feature term combination is calculated. The support score of each candidate set is used to filter out... The candidate set ≥ min_sup constitutes the frequent feature itemset. .
[0042] For example, in In this case, connect each pair of features to generate a combination of candidate features. If support({ If})=0.12≥0.1, then { }∈ .
[0043] Next, from the frequent feature itemset The feature combination with a confidence level not less than the preset confidence threshold min_conf is extracted and determined as the associated feature. Specifically, for any two disjoint feature subsets... X and Y , where X Y X∩Y= The confidence level is calculated using the following formula: ; like When the value is greater than or equal to min_conf, then the corresponding feature combination... It has been identified as a related feature.
[0044] For example, in support({ })=0.12, =0.15, and with the preset confidence threshold min_conf set to 0.7, the confidence level is... =0.8≥0.7, therefore, the feature combination { } was identified as a related feature.
[0045] Therefore, by fusing the comprehensive feature values and the associated features, a fused feature dataset is obtained.
[0046] S104: Based on the fused feature dataset, generate a set of user tags and construct a user profile.
[0047] In this embodiment, based on the fused feature dataset, multiple tags included in the preset tag library are scored to obtain a score value corresponding to each tag. Specifically, this embodiment can use a tag scoring algorithm, such as the random forest algorithm, to calculate the matching degree score between the user and each tag in the preset tag library based on the fused feature dataset, thereby obtaining a score value corresponding to each tag, and filtering out tags with score values greater than a preset score threshold to obtain an initial tag set.
[0048] It should be noted that the preset tag library includes multiple tags covering major categories such as demographic attributes, interests and preferences, consumption potential, and scenario needs. These tags can comprehensively cover multi-dimensional user characteristics, providing comprehensive tag support for subsequent tag scoring and user profile construction.
[0049] Next, the rating values corresponding to each tag in the initial tag set are updated based on the time decay factor, resulting in updated rating values for each tag. The time decay factor can be set based on the time freshness of user behavior, decaying historical tag rating values to obtain updated rating values for each tag. Tags with updated rating values greater than a preset rating threshold are filtered to obtain a user tag set. Based on this user tag set, a user profile is constructed.
[0050] Meanwhile, this application embodiment also updates the tag score based on the newly generated behavioral data, and realizes dynamic iteration of tags. For example, when a user frequently visits parenting apps recently, the tag "parent-child group" is automatically added.
[0051] After constructing user profiles based on user tag sets, scenario-based profile adaptation and optimization are necessary to better adapt these profiles to actual business needs. Specifically, based on specific business scenarios (such as targeted marketing or data plan recommendations), a scenario feature weight model is invoked to adjust the weight ratio of different tags. This scenario feature weight model pre-defines core feature dimensions for different scenarios. For example, the data plan recommendation scenario emphasizes the weight of tags such as data usage and data consumption from frequently accessed apps, while the local life marketing scenario emphasizes the weight of tags such as base station activity areas and access to local service mini-programs. Through weight adjustment, the optimized scenario-based user profile is finally output. This profile includes core tags, feature weights, and behavioral interpretations, and can be directly adapted to the actual needs of various business scenarios.
[0052] After completing the adaptation and optimization of scenario-based user profiles, this application embodiment constructs a feedback loop and model optimization mechanism to further improve the accuracy of user profiles and related models. Specifically, it collects business feedback data after the application of scenario-based user profiles (such as marketing campaign conversion rates and user acceptance of recommendation services), and combines this feedback with evaluation opinions from domain experts, feeding the aforementioned feedback information back into the tag scoring model and scenario feature weight model mentioned above. On the one hand, it optimizes the accuracy of tag matching by adjusting model parameters; on the other hand, it updates the threshold and weight configuration of associated feature mining to improve the effectiveness of feature fusion.
[0053] This application's embodiments achieve dynamic tag iteration by setting a time decay factor to attenuate historical tag scores and updating features and recalculating tag scores based on newly generated behavioral data. This effectively solves the problem in existing technologies where tags are mostly preset static tags, lacking a dynamic update mechanism, and failing to adapt to real-time changes in user behavior preferences, resulting in poor timeliness of user profiles. Simultaneously, by constructing a scenario-based profile adaptation and optimization mechanism, it combines specific business scenarios with scenario feature weight models to adjust tag weight ratios, outputting scenario-based user profiles adapted to various business scenarios. This addresses the issue that profile results are not optimized for specific business scenarios, making it difficult to directly support actual operational needs such as personalized service recommendations and precision marketing.
[0054] Based on the relevant content of steps S101-S104 above, it can be seen that: First, multi-source user behavior data is collected, including communication-side data and business application-side data. Then, features are extracted from the multi-source user behavior data to generate an initial feature dataset. Next, the multi-source features in the initial feature dataset are fused to obtain a fused feature dataset. Finally, based on the fused feature dataset, a user tag set is generated and a user profile is constructed. It is evident that this solution collects multi-source user behavior data, such as communication-side data and business application-side data, extracts features from it to generate an initial feature dataset, then fuses the multi-source features in the initial feature dataset to obtain a fused feature dataset, and finally generates user tags and constructs a user profile based on the fused feature dataset. Through this multi-source feature fusion method, this solution can effectively integrate data from different sources, overcome the data fragmentation problem in existing technologies, and enable the constructed user profile to more comprehensively and accurately reflect user behavior characteristics and preferences, thereby improving the accuracy and reliability of the user profile.
[0055] It should be noted that the user information (including but not limited to basic information such as user location, name, age and ID number) and data (including but not limited to user location data and user behavior data, as well as data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.
[0056] Furthermore, the figure is a schematic diagram of a user profile building device provided in an embodiment of this application. (Combined with...) Figure 2 As shown, the user profile building apparatus 200 provided in this application embodiment may include: Data acquisition module 201 is used to collect multi-source user behavior data, which includes communication side data and business application side data; Feature extraction module 202 is used to extract features from the multi-source user behavior data and generate an initial feature dataset; Feature fusion module 203 is used to fuse the multi-source features in the initial feature dataset to obtain a fused feature dataset; The profile building module 204 is used to generate a set of user tags and build a user profile based on the fused feature dataset.
[0057] Optionally, the feature fusion module 203 may include: The feature weighting module is used to perform weighted summation on the multi-source features in the initial feature dataset to obtain a comprehensive feature value; The feature mining module is used to perform association rule mining on the multi-source features in the initial feature dataset to obtain associated features; The fusion processing module is used to fuse the comprehensive feature values and the associated features to obtain a fused feature dataset.
[0058] Optionally, the feature weighting module may include: The weight determination module is used to determine the credibility weight of each feature in the multi-source features based on the data stability and data integrity of each feature. The standardization processing module is used to standardize the observed values of each feature in the multi-source features to determine the standard feature value of each feature in the multi-source features; The weighted summation module is used to perform a weighted summation of the standard feature values of each feature in the multi-source features based on the confidence weight of each feature in the multi-source features, so as to obtain the comprehensive feature value corresponding to the multi-source features.
[0059] Optionally, the feature mining module may include: A transaction set construction module is used to construct a transaction set based on the multi-source features in the initial feature dataset, wherein the transaction set includes multiple feature items of the user; The frequent feature set determination module is used to mine multiple feature items in the transaction set based on a preset mining algorithm to determine the frequent feature item set in the transaction set. The association feature determination module is used to extract feature item combinations with a confidence level not less than a preset confidence threshold from the set of frequent feature items and determine them as association features.
[0060] Optionally, the weight determination module is specifically used for: Obtain the observation value and data record of each feature in the multi-source features within the historical time window; Based on the standard deviation of the observed values, the stability score of the data source corresponding to each feature is determined; Based on the data records, the number of historical time windows without missing values is counted, and the ratio of the number of historical time windows without missing values to the total number of historical time windows is determined as the integrity score of the data source corresponding to each feature. The stability score and the integrity score are weighted and summed to determine the credibility weight of each feature in the multi-source features.
[0061] Optionally, the association feature determination module is specifically used for: Based on the set of frequent feature terms, candidate feature term combinations are generated; Based on the candidate feature combination, the confidence level of the candidate feature combination is calculated; The feature items with a confidence level not less than a preset confidence threshold are combined and determined as associated features.
[0062] Optionally, the image construction module 204 is specifically used for: Based on the fused feature dataset, multiple tags included in the preset tag library are scored to obtain the score value corresponding to each tag; The tags with scores greater than a preset score threshold are filtered to obtain an initial tag set; The score values corresponding to each tag in the initial tag set are updated based on the time decay factor to obtain the updated score values of each tag; The updated rating values of the tags that are greater than the preset rating threshold are filtered to obtain a set of user tags. Based on the user tag set, the user profile is constructed. Furthermore, embodiments of this application also provide an electronic device, including: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, the one or more programs including instructions, which, when executed by the processor, cause the processor to perform any of the implementation steps of the user profile construction method described above.
[0063] Furthermore, embodiments of this application also provide a computer-readable storage medium for storing a computer program, which, when executed by a terminal device, implements any of the implementation steps of the user profile construction method described above.
[0064] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application. It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on describing the differences from other embodiments. The same or similar parts between the various embodiments can be referred to mutually.
[0065] The system disclosed in the embodiments is described simply because it corresponds to the method disclosed in the embodiments; relevant details can be found in the method section.
[0066] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0067] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for constructing user profiles, characterized in that, The method includes: Collect multi-source user behavior data, which includes communication-side data and business application-side data; Feature extraction is performed on the multi-source user behavior data to generate an initial feature dataset; The multi-source features in the initial feature dataset are fused to obtain a fused feature dataset; Based on the fused feature dataset, a set of user tags is generated and a user profile is constructed.
2. The method according to claim 1, characterized in that, The process of fusing multi-source features in the initial feature dataset to obtain a fused feature dataset includes: The multi-source features in the initial feature dataset are weighted and summed to obtain a comprehensive feature value; The multi-source features in the initial feature dataset are subjected to association rule mining to obtain associated features; The integrated feature values and the associated features are fused together to obtain a fused feature dataset.
3. The method according to claim 2, characterized in that, The step of performing a weighted summation of the multi-source features in the initial feature dataset to obtain comprehensive features includes: Based on the data stability and data integrity of each feature in the multi-source features, the credibility weight of each feature in the multi-source features is determined; The observed values of each feature in the multi-source features are standardized to determine the standard feature value of each feature in the multi-source features; Based on the credibility weight of each feature in the multi-source features, the standard feature values of each feature in the multi-source features are weighted and summed to obtain the comprehensive feature value corresponding to the multi-source features.
4. The method according to claim 2, characterized in that, The step of performing association rule mining on the multi-source features in the initial feature dataset to obtain association features includes: Based on the multi-source features in the initial feature dataset, a transaction set is constructed, which includes multiple feature items of the user. Based on a preset mining algorithm, multiple feature items in the transaction set are mined to determine the frequent feature item set in the transaction set; The combination of feature terms with a confidence level not less than a preset confidence threshold is extracted from the set of frequent feature terms and determined as associated features.
5. The method according to claim 3, characterized in that, The determination of the credibility weight of each feature in the multi-source features based on the data stability and data integrity of each feature includes: Obtain the observation value and data record of each feature in the multi-source features within the historical time window; Based on the standard deviation of the observed values, the stability score of the data source corresponding to each feature is determined; Based on the data records, the number of historical time windows without missing values is counted, and the ratio of the number of historical time windows without missing values to the total number of historical time windows is determined as the integrity score of the data source corresponding to each feature. The stability score and the integrity score are weighted and summed to determine the credibility weight of each feature in the multi-source features.
6. The method according to claim 4, characterized in that, The step of extracting feature item combinations with a confidence level not less than a preset confidence threshold from the set of frequent feature items and determining them as associated features includes: Based on the set of frequent feature terms, candidate feature term combinations are generated; Based on the candidate feature combination, the confidence level of the candidate feature combination is calculated; The feature items with a confidence level not less than a preset confidence threshold are combined and determined as associated features.
7. The method according to claim 1, characterized in that, The process of generating a user tag set and constructing a user profile based on the fused feature dataset includes: Based on the fused feature dataset, multiple tags included in the preset tag library are scored to obtain the score value corresponding to each tag; The tags with scores greater than a preset score threshold are filtered to obtain an initial tag set; The score values corresponding to each tag in the initial tag set are updated based on the time decay factor to obtain the updated score values of each tag; The updated rating values of the tags that are greater than the preset rating threshold are filtered to obtain a set of user tags. The user profile is constructed based on the user tag set.
8. A user profile building device, characterized in that, include: The data acquisition module is used to collect multi-source user behavior data, which includes communication-side data and business application-side data. The feature extraction module is used to extract features from the multi-source user behavior data and generate an initial feature dataset; The feature fusion module is used to fuse the multi-source features in the initial feature dataset to obtain a fused feature dataset; The profile building module is used to generate a set of user tags and build user profiles based on the fused feature dataset.
9. An electronic device, characterized in that, The device includes: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store a program, the program including instructions that, when executed by the processor, cause the processor to perform the steps of the user profile construction method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, which, when executed by a terminal device, implements the steps of the user profile construction method according to any one of claims 1-7.