Computer cloud data optimization method and system based on deep learning

Abstract

The present application discloses a computer cloud data optimization method and system based on deep learning, which relates to the field of big data technology, including: S1, obtaining behavior preference information of interactive big data, and generating a behavior sequence of the behavior preference information; S2, based on the obtained behavior sequence, dividing the behavior preference information into multiple data sets, and outputting them as a first target set; S3, determining the exploration step lengths corresponding to different pages under each first target set, marking the exploration step lengths according to different user demand types, and outputting them as a second target set; S4, determining the compensation coefficient corresponding to the user behavior based on the obtained first target set and the second target set, and determining the behavior characteristics corresponding to the user behavior based on the compensation coefficient; S5, determining the calculated behavior recommendation value based on the user's behavior characteristics, and outputting the corresponding recommendation optimization result; it can improve the analysis effect of user behavior and improve the accuracy of recommendation.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and in particular to a computer cloud data optimization method and system based on deep learning. Background Art

[0002] In the process of continuous social and economic development, the amount of information data has also shown a rapid growth trend. At the same time, the storage, integration, and search of information data have further increased the requirements for computer network technology. Therefore, artificial intelligence has become the most effective way to help people process information data. The application of artificial intelligence technology in computer networks in the era of big data is an efficient and accurate network technology. It can use its own intelligent advantages to reasonably classify and process massive data information in computer systems. At present, data mining and analysis technology based on artificial intelligence is still popular, and how to achieve high-quality data optimization processing at the feature level is one of the difficulties.

[0003] For example, Chinese patent publication No. CN115982236A discloses a big data optimization method and server for AI. By aggregating behavior preference description knowledge KV2 and behavior preference description knowledge KV3 to obtain behavior preference description knowledge KV4, the quality of optimizing the user behavior preference vector contained in behavior preference description knowledge KV2 using the user behavior preference vector contained in behavior preference description knowledge KV1 can be improved, that is, the data details of optimizing the second digital service interaction big data using the user behavior preference vector of the first digital service interaction big data can be improved.

[0004] However, when optimizing the corresponding behavioral preferences, it is too biased towards explaining multiple steps and dimensions, with poor interpretability of the specific implementation methods. In addition, when processing the corresponding behavioral preferences, it does not take into account the differences in usage by different users, resulting in step limitations and prediction value deviations in the final results. Summary of the invention

[0005] The embodiments of the present application provide a computer cloud data optimization method and system based on deep learning, which solves the technical problem of how to effectively extract user behavior preferences from interactive big data, further classify and optimize user behavior data, and provide users with personalized recommendation results based on this.

[0006] The embodiment of the present application provides a computer cloud data optimization method based on deep learning, comprising: S1, obtaining behavior preference information of interactive big data, and generating a behavior sequence of the behavior preference information;

[0007] S2, based on the acquired behavior sequence, dividing the behavior preference information into multiple data sets, and outputting them as a first target set;

[0008] S3, determining the exploration steps corresponding to different pages under each first target set, marking the exploration steps according to different user demand types, and outputting them as a second target set;

[0009] S4, determining a compensation coefficient corresponding to the user behavior based on the acquired first target set and second target set, and determining a behavior feature corresponding to the user behavior based on the compensation coefficient;

[0010] S5, based on the user's behavior characteristics, determine the calculated behavior recommendation value and output the corresponding recommendation optimization result.

[0011] Step S3 also includes, S31, determining the user's behavior pattern based on the acquired first target set, the behavior pattern including acquiring the user's stay time on the page, click frequency, and jump path;

[0012] S32, determining the exploration step lengths corresponding to different pages based on the user's behavior pattern;

[0013] S33, based on the user's demand type, adjusting the exploration step lengths of different pages, and outputting the data whose exploration step lengths meet the preset conditions as the second target set.

[0014] Step S4 also includes, S41, judging a behavior state corresponding to a current user behavior based on the first target set and the second target set, and calculating a current compensation coefficient based on the behavior state;

[0015] S42, performing real-time compensation on the user behavior according to the current compensation coefficient to obtain a current effective compensation value;

[0016] S43, adjusting the current compensation coefficient based on the effective compensation value, and outputting the adjusted compensation coefficient;

[0017] S44, determining output behavior characteristics according to the adjusted compensation coefficient.

[0018] A computer cloud data optimization system based on deep learning, including a data acquisition module for acquiring behavior preference information corresponding to user behavior and generating a corresponding behavior sequence;

[0019] A first processing module, configured to output a first target set according to the acquired behavior preference information;

[0020] A second processing module is used to mark the first target set according to the acquired first target set, and output a second target set;

[0021] A feature extraction module, used to determine a compensation coefficient of the user behavior based on the first target set and the second target set, and output a behavior feature corresponding to the compensation coefficient;

[0022] The recommendation optimization module is used to determine the corresponding behavior recommendation value according to the obtained behavior characteristics and generate the recommendation optimization result.

[0023] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0024] By acquiring user behavior data and generating behavior sequences; using the first processing module to classify these behavior sequences to form a first target set; further analyzing the user's behavior pattern, such as dwell time, click frequency, jump path, etc., through the second processing module, and determining the exploration step between different pages to form a second target set; combining the information of the first target set and the second target set, determining the compensation coefficient of the user behavior, and then extracting key user behavior features; using the extracted behavior features to calculate the recommendation value, and optimizing the recommendation results through methods such as weighted averaging to more accurately meet the user's personalized needs; being able to efficiently process and analyze large-scale user behavior data; by analyzing the user's behavior pattern and exploration step, the system can have a deeper understanding of the user's preferences and needs; through the introduction of the compensation coefficient and the extraction of behavior features, being able to more accurately provide users with personalized recommendation content and improve user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a flowchart of a computer cloud data optimization method based on deep learning of the present invention;

[0026] Figure 2 This is a flow chart of step S3 of the computer cloud data optimization method based on deep learning of the present invention;

[0027] Figure 3 This is a flow chart of step S4 of the computer cloud data optimization method based on deep learning of the present invention;

[0028] Figure 4 This is a system schematic diagram of the computer cloud data optimization system based on deep learning of the present invention. DETAILED DESCRIPTION

[0029] To facilitate the understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0030] It should be noted that the terms “vertical”, “horizontal”, “up”, “down”, “left”, “right” and similar expressions used in this document are only for illustrative purposes and do not represent the only implementation method.

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which the present invention belongs; the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more related listed items. Embodiment 1

[0032] like Figure 1 As shown, the computer cloud data optimization method based on deep learning in this application includes:

[0033] S1, obtain the behavior preference information of the interactive big data and generate the behavior sequence of the behavior preference information;

[0034] S2, based on the acquired behavior sequence, dividing the behavior preference information into multiple data sets, and outputting them as a first target set;

[0035] S3, determining the exploration steps corresponding to different pages under each first target set, marking the exploration steps according to different user demand types, and outputting them as a second target set;

[0036] S4, determining a compensation coefficient corresponding to the user behavior based on the acquired first target set and second target set, and determining a behavior feature corresponding to the user behavior based on the compensation coefficient;

[0037] S5, based on the user's behavior characteristics, determine the calculated behavior recommendation value and output the corresponding recommendation optimization result.

[0038] Among them, the behavioral preference information obtained from the interactive big data includes user historical behavior data, user portraits, and user behavior sequences in a single session, such as page browsing order, dwell time, and the depth of the user's page (the number of clicks from the access entry point to the current page), to obtain the user's behavioral preferences when making queries; the exploration step corresponding to different pages refers to the number of clicks or path changes required to reach the target browsed web page during page browsing; the compensation coefficient refers to the compensation number corresponding to the loss value between page browsing and behavior sequence, to make the calculated prediction value closer to the user's habitual value.

[0039] By obtaining the user's basic information, device information, and social network information, and based on the user's corresponding page operations when obtaining behavioral information, the difference between the user's operation behavior and the actual query content is determined, thereby improving the efficiency and accuracy of processing user behavior characteristics. Embodiment 2

[0040] In order to sort the obtained behavior sequence according to the page browsing order, dwell time, and the depth of the page where the user is located, the obtained behavior sequence is classified according to different numerical values, thereby obtaining a classification effect for the current interface.

[0041] Specifically, the implementation of step S2 also includes:

[0042] Based on the acquired behavior sequence, the corresponding behavior preference information under each behavior sequence is determined, the acquired behavior preference information is classified according to the user's page browsing order, the user's residence time under each category and the user's browsing depth under the corresponding keyword are recorded, the behavior preference information under each category is sorted according to the values ​​of page browsing order, residence time, and page depth, and output in sequence as the first target set.

[0043] By processing the acquired behavior sequence according to the different browsing orders and depths of users and obtaining the first target set related to browsing, the weight and preference of each browsing interface when the user browses can be identified, thereby assisting in the identification of user behavior and improving the analysis effect of user behavior. Embodiment 3

[0044] After obtaining the first target set, the first target set is processed, the depth of each actual browsing interface is calculated, the number of times the user jumps between interfaces with different tags, and the changes in the corresponding browsing tags when the interface jumps are recorded, and the changes in the tags are recorded as the exploration step length, and the user's browsing behavior is reclassified according to the exploration step length to obtain the classified second target set. Among them, the exploration step length can be understood as the span or frequency of the user jumping from one page to another during the browsing or search process. It reflects the speed and breadth of the user's exploration of information and is an important parameter for optimizing the user's search experience and recommendation effect.

[0045] Specifically, Figure 2 As shown, the implementation of step S3 also includes:

[0046] S31, based on the obtained first target set, determining the user's behavior pattern, the behavior pattern includes obtaining the user's stay time on the page, click frequency, and jump path.

[0047] After obtaining the first target set, the data in the first target set is identified according to the obtained first target set, and the values ​​corresponding to the user's stay time, click frequency, and jump path in the first target set are determined, thereby determining the user's span between interfaces and data recommendation parameters.

[0048] Statistical analysis and data mining techniques are used to analyze user behavior patterns on the page. For example, the average dwell time on each page, click-through rate (number of clicks / number of page impressions), and frequent patterns of jump paths can be calculated.

[0049] S32, based on the user's behavior pattern, determine the exploration step lengths corresponding to different pages; for example, for pages that the user often quickly skips, a smaller exploration step length can be set; and for pages that the user often browses in depth, a larger exploration step length can be set.

[0050] The implementation of step S32 further includes:

[0051] Obtain the average number of jumps and time intervals of users from one page to another during browsing, determine the jump frequency corresponding to the browsing interface, and determine the exploration step corresponding to the current user behavior based on the obtained jump frequency.

[0052] S33, based on the user's demand type, adjusting the exploration step lengths of different pages, and outputting the data whose exploration step lengths meet the preset conditions as the second target set.

[0053] Among them, the preset conditions include matching the current user behavior based on the acquired exploration step, determining the matching method between the exploration step and the user behavior, and matching the user behavior with the exploration step according to the corresponding behavior preference information and behavior type at the time of jump, and determining the corresponding behavior matching result after matching, wherein the behavior matching result includes one or more demand types, each demand type corresponds to one or more exploration steps, until all exploration steps are matched with the user behavior, and the matched behavior matching result is output as the second target set.

[0054] Preferably, when the exploration step length matches the user's demand type, it also includes outputting the acquired exploration step length to the deep learning network to obtain the target matching behavior, and based on the target matching behavior, determining a first matching relationship between the target matching behavior and the behavior preference information and a second matching relationship between the target matching behavior and the behavior to be identified, and outputting the behavior matching result based on the acquired first matching relationship and second matching relationship.

[0055] Among them, the first matching relationship describes the similarity or consistency between the target matching behavior obtained by the deep learning network and the user's original behavior preference information, and the second matching relationship refers to the similarity or correlation between the target matching behavior obtained by the deep learning network and the user behavior that currently needs to be identified (i.e., the behavior to be identified).

[0056] Preferably, when determining the behavior matching results, for each target matching behavior, determine whether the user behaviors corresponding to the first matching relationship and the second matching relationship are consistent, and for each target matching behavior, compare the similarity score of the first matching relationship and the second matching relationship with the set threshold; when the similarity score of the first matching relationship and the second matching relationship is higher than the set threshold, output the target matching behavior that is higher than the set threshold as the second target set.

[0057] If the similarity scores of two matching relationships are both higher than their respective thresholds, it can be considered that the user's behavior is consistent, that is, the target matching behavior is highly matched with the user's original behavior preference information and the current behavior to be identified; if the similarity score of any matching relationship is lower than the threshold, it may indicate that the user's behavior has changed or is abnormal, and further analysis or other measures are needed. Embodiment 4

[0058] After obtaining and processing the actual browsing data according to the user class, the first target set obtained is compared with the second target set, and the same preferences that are likely to appear in the two sets are screened out to determine the behavioral characteristics that are likely to be generated according to different choices when the user browses, and identify the data related to the corresponding behavioral characteristics to determine the behavioral associations for different users.

[0059] Specifically, Figure 3 As shown, the specific implementation of step S4 includes:

[0060] S41, based on the first target set and the second target set, determining a behavior state corresponding to the current user behavior, and calculating a current compensation coefficient based on the behavior state;

[0061] S42, performing real-time compensation on the user behavior according to the current compensation coefficient to obtain a current effective compensation value;

[0062] S43, adjusting the current compensation coefficient based on the effective compensation value, and outputting the adjusted compensation coefficient;

[0063] S44, determining output behavior characteristics according to the adjusted compensation coefficient.

[0064] Among them, the behavior state refers to the specific mode or state of the user when searching, browsing or other interactive behaviors. These states can reflect the user's interest in the current content, satisfaction, willingness to explore, etc. For example, the user may be in different behavior states such as "highly focused" (deep exploration of a topic), "browsing" (extensively but not deeply viewing multiple topics), or "lost" (unable to find content of interest).

[0065] The method for determining the compensation coefficient includes determining the key factors of user behavior, such as the price discount of the product, the user's preference for the brand, the popularity of the product, etc., based on the first target set and the second target set; setting a basic score for the user behavior, identifying the user's current behavior state based on the basic score and key factors of the user behavior, and defining one or more compensation coefficients for each user behavior according to the user's behavior state and key factors.

[0066] Assume there is a product A, whose base score is base_A, and it is identified that the user has a preference for the brand to which the product belongs, and the product is currently discounted. Two compensation coefficients can be defined: brand_preference_coefficient and discount_coefficient.

[0067] The calculation formula of the compensation coefficient (final_A) can be:

[0068] final_A=base_A×(1+brand_preference_coefficient)×(1+discount_coefficient)

[0069] Among them, base_A is the basic score of product A, brand_preference_coefficient is the compensation coefficient calculated based on the user's preference for the brand. If the user has a high preference for the brand, this coefficient may be positive; if the user has no preference or a low preference for the brand, this coefficient may be zero or negative, and discount_coefficient is the compensation coefficient calculated based on the discount strength of the product. The greater the discount strength, the higher the coefficient.

[0070] In this way, the order of the recommendation list can be adjusted according to the user's real-time behavior status and the dynamic attributes of the product, thereby improving user satisfaction and purchase conversion rate. It should be noted that the specific definition and calculation formula of the compensation coefficient will vary according to different business scenarios and needs.

[0071] After obtaining the user's behavior state, the processing method for the behavior state also includes calculating the difference between the compensation coefficients obtained twice adjacently. If the absolute value of the difference between the compensation coefficients is less than the change threshold, the count value is increased by 1, otherwise the count value is cleared; if the count value continues to accumulate to a number greater than the threshold, the current behavior state is in a stable state, and then the count value is cleared; when the absolute value of the difference between the compensation coefficients is greater than the change threshold, the current behavior state is an unstable state.

[0072] When the count value is reset, the compensation coefficient at the previous moment is used as the compensation coefficient of the current behavior state, and the effective compensation value of the current behavior state is adjusted.

[0073] The effective compensation value obtained is expressed as follows:

[0074] valid=raw-ref×final_A

[0075] Among them, valid is the current valid compensation value, raw is the current initial value, ref is the current reference compensation value, and final_A is the compensation coefficient.

[0076] If the adjusted effective compensation value is greater than the maximum compensation value, the maximum compensation coefficient is used as the current compensation coefficient. If the adjusted effective compensation value is less than the minimum compensation value, the minimum compensation coefficient is used as the current compensation coefficient. If the adjusted effective compensation value is less than the maximum compensation value and greater than the minimum compensation value, the compensation coefficient corresponding to the adjusted effective compensation value is used as the current compensation coefficient.

[0077] According to the adjusted compensation coefficient, the corresponding user behavior is obtained, and the corresponding behavior characteristics are output.

[0078] The compensation coefficient may be a weight value calculated based on multiple factors such as user behavior characteristics, page importance, user demand type, etc., and is used to adjust the sorting and display of search results.

[0079] Combined with the compensation coefficient, the behavioral characteristics of users during search are further determined. These behavioral characteristics may include the selection of search keywords, the click-through rate of search results, the dwell time on the search page, etc., which reflect the preferences and needs of users during the search process. Embodiment 5

[0080] After obtaining the corresponding processed behavioral features, the obtained behavioral features are extracted to determine the recommendation value of the corresponding behavior in the process of each user's behavior, and the user's interest level and recommendation priority for a certain interface or content are determined based on these recommendation values, and the recommendation value is output to improve the efficiency of user behavior analysis and reduce the complexity of calculation.

[0081] Specifically, the behavioral characteristics obtained in step S5 may include user preference characteristics: such as the user's preference for different categories of content, brand preference, etc.; user activity characteristics: such as the user's login frequency, browsing time, click-through rate, etc.; user consumption capacity characteristics: such as the user's purchase frequency, purchase amount, purchased product category, etc.; time series characteristics: such as the user's active time period in a day or a week, seasonal purchase patterns, etc.

[0082] Based on the extracted user behavior features, a suitable algorithm or model is used to calculate the behavior recommendation value. This recommendation value can be a score, ranking, or probability, indicating the strength or priority of the system's recommendation to the user for a specific behavior (such as purchasing a certain product or watching a certain video).

[0083] Methods for calculating behavior recommendation values ​​may include:

[0084] Based on the obtained behavior characteristics and the first target set, determine the first target content corresponding to the user behavior and the behavior of similar users; based on the obtained behavior characteristics, the first target set, and the second target set, determine the second target content related to the characteristics of the user behavior itself and the user's historical preferences; input the obtained first target content and the second target content into the convolutional neural network, and output the calculated behavior recommendation value together with the corresponding behavior characteristics. This can be a list, matrix, or database query result, which contains the feature description and recommendation value of each recommended behavior.

[0085] After obtaining the behavior recommendation value, the product of the recommendation value and the compensation coefficient is weighted and averaged to output the final recommendation optimization result.

[0086] like Figure 4 As shown, the present application also provides a computer cloud data optimization system based on deep learning, including:

[0087] A data acquisition module is used to obtain behavior preference information corresponding to user behavior and generate a corresponding behavior sequence;

[0088] A first processing module, configured to output a first target set according to the acquired behavior preference information;

[0089] A second processing module is used to mark the first target set according to the acquired first target set, and output a second target set;

[0090] A feature extraction module, used to determine a compensation coefficient of the user behavior based on the first target set and the second target set, and output a behavior feature corresponding to the compensation coefficient;

[0091] The recommendation optimization module is used to determine the corresponding behavior recommendation value according to the obtained behavior characteristics and generate the recommendation optimization result.

[0092] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A computer cloud data optimization method based on deep learning, It is characterized in that include: S1, obtain the behavior preference information of the interactive big data and generate the behavior sequence of the behavior preference information; S2, based on the acquired behavior sequence, dividing the behavior preference information into multiple data sets, and outputting them as a first target set; S3, determining the exploration steps corresponding to different pages under each first target set, marking the exploration steps according to different user demand types, and outputting them as a second target set; S4, determining a compensation coefficient corresponding to the user behavior based on the acquired first target set and second target set, and determining a behavior feature corresponding to the user behavior based on the compensation coefficient; S5, based on the user's behavior characteristics, determine the calculated behavior recommendation value and output the corresponding recommendation optimization result; Step S3 also includes, S31, determining the user's behavior pattern based on the acquired first target set, the behavior pattern including acquiring the user's stay time on the page, click frequency, and jump path; S32, determining the exploration step lengths corresponding to different pages based on the user's behavior pattern; S33, adjusting the exploration step lengths of different pages based on the user's demand type, and outputting data whose exploration step lengths meet preset conditions as a second target set; The preset conditions include matching the current user behavior based on the acquired exploration step length, determining the matching method between the exploration step length and the user behavior, and matching the user behavior with the exploration step length according to the corresponding behavior preference information and behavior type at the time of jump, and determining the corresponding behavior matching result after matching, wherein the behavior matching result includes one or more demand types, and each demand type corresponds to one or more exploration step lengths, until all the exploration step lengths are matched with the user behavior, and the matched behavior matching result is output as the second target set; Step S4 also includes, S41, judging a behavior state corresponding to a current user behavior based on the first target set and the second target set, and calculating a current compensation coefficient based on the behavior state; S42, performing real-time compensation on the user behavior according to the current compensation coefficient to obtain a current effective compensation value; S43, adjusting the current compensation coefficient based on the effective compensation value, and outputting the adjusted compensation coefficient; S44, determining output behavior characteristics according to the adjusted compensation coefficient.

2. The computer cloud data optimization method based on deep learning as claimed in claim 1, It is characterized in that Based on the acquired behavior sequence, the corresponding behavior preference information under each behavior sequence is determined, the acquired behavior preference information is classified according to the user's page browsing order, the user's residence time under each category and the user's browsing depth under the corresponding keyword are recorded, the behavior preference information under each category is sorted according to the values ​​of page browsing order, residence time, and page depth, and output in sequence as the first target set.

3. The computer cloud data optimization method based on deep learning as claimed in claim 1, It is characterized in that When the exploration step length matches the user's demand type, it also includes outputting the acquired exploration step length to the deep learning network to obtain the target matching behavior, and based on the target matching behavior, determining a first matching relationship between the target matching behavior and the behavior preference information and a second matching relationship between the target matching behavior and the behavior to be identified, and outputting the behavior matching result based on the acquired first matching relationship and second matching relationship.

4. The computer cloud data optimization method based on deep learning as claimed in claim 3, It is characterized in that When determining the behavior matching results, for each target matching behavior, determine whether the user behaviors corresponding to the first matching relationship and the second matching relationship are consistent. For each target matching behavior, compare the similarity score of the first matching relationship and the second matching relationship with the set threshold; when the similarity score of the first matching relationship and the second matching relationship is higher than the set threshold, output the target matching behavior that is higher than the set threshold as the second target set.

5. The computer cloud data optimization method based on deep learning as claimed in claim 1, It is characterized in that If the adjusted effective compensation value is greater than the maximum compensation value, the maximum compensation coefficient is used as the current compensation coefficient. If the adjusted effective compensation value is less than the minimum compensation value, the minimum compensation coefficient is used as the current compensation coefficient. If the adjusted effective compensation value is less than the maximum compensation value and greater than the minimum compensation value, the compensation coefficient corresponding to the adjusted effective compensation value is used as the current compensation coefficient.

6. The computer cloud data optimization method based on deep learning as claimed in claim 1, It is characterized in that Methods for calculating behavior recommendation values ​​include: Based on the obtained behavior characteristics and the first target set, determine the first target content corresponding to the user behavior and the behavior of similar users; based on the obtained behavior characteristics, the first target set, and the second target set, determine the second target content related to the characteristics of the user behavior itself and the user's historical preferences; input the obtained first target content and the second target content into the convolutional neural network, and output the calculated behavior recommendation value together with the corresponding behavior characteristics; After obtaining the behavior recommendation value, the product of the recommendation value and the compensation coefficient is weighted and averaged to output the final recommendation optimization result.

7. A computer cloud data optimization system based on deep learning, using the computer cloud data optimization method based on deep learning as claimed in claim 1, It is characterized in that It includes a data acquisition module, which is used to obtain behavior preference information corresponding to user behavior and generate a corresponding behavior sequence; A first processing module, configured to output a first target set according to the acquired behavior preference information; A second processing module is used to mark the first target set according to the acquired first target set, and output a second target set; Among them, based on the obtained first target set, the user's behavior pattern is determined, and based on the user's behavior pattern, the behavior pattern includes obtaining the user's residence time on the page, click frequency, and jump path; the exploration step corresponding to different pages is determined, and the exploration step of different pages is adjusted based on the user's demand type, and the data whose exploration step meets the preset conditions is output as the second target set; the preset conditions include, based on the obtained exploration step, matching the current user behavior, determining the matching method between the exploration step and the user behavior, and matching the user behavior with the exploration step according to the corresponding behavior preference information and behavior type at the time of jumping, and determining the corresponding behavior matching result after matching, wherein the behavior matching result includes one or more demand types, and each demand type corresponds to one or more exploration steps, until all exploration steps are matched with the user behavior, and the matched behavior matching result is output as the second target set; A feature extraction module, used to determine a compensation coefficient of the user behavior based on the first target set and the second target set, and output a behavior feature corresponding to the compensation coefficient; Wherein, based on the first target set and the second target set, the behavior state corresponding to the current user behavior is determined, and based on the behavior state, the current compensation coefficient is calculated; based on the current compensation coefficient, the user behavior is compensated in real time to obtain the current effective compensation value; based on the effective compensation value, the current compensation coefficient is adjusted, and the adjusted compensation coefficient is output; based on the adjusted compensation coefficient, the output behavior feature is determined; The recommendation optimization module is used to determine the corresponding behavior recommendation value according to the obtained behavior characteristics and generate the recommendation optimization result.

Images