Big data driven intelligent recommendation system

Abstract

The invention relates to the technical field of big data processing and intelligent recommendation, in particular to an intelligent recommendation system driven by big data. The method comprises the following steps: collecting user behavior data and article feature data, and generating a user feature set and an article feature set; a user feature weight is introduced, a relative deviation degree is calculated, a recommendation value is calculated, and a recommendation list is generated; displaying the recommendation list to the user to obtain a feedback value; and dynamically adjusting the user feature weight based on the feedback value, the recommendation value and the relative deviation degree, calculating the relative deviation degree and the recommendation value based on the obtained final user feature weight, and finally generating a personalized recommendation list. The problem that a traditional recommendation system mainly depends on static similarity calculation of users and articles, so that personalized recommendation accuracy is insufficient is solved; the response capability to the user interest change is weak, so that the recommendation system easily misjudges the user demand; and a global performance optimization and regulation and control mechanism is lacked, so that a recommendation system is easy to fall into local optimum.

Description

Technical Field

[0001] The present invention relates to the technical field of big data processing and intelligent recommendation, and in particular to a big data driven intelligent recommendation system. Background Art

[0002] At present, recommendation systems have been widely used in e-commerce, social networks and content platforms. Traditional recommendation systems mainly rely on content-based and collaborative filtering algorithms to recommend products that match user interests based on the similarity between user historical behavior and item features. However, with the continuous growth of data scale and the diversification of user interests, traditional recommendation systems have shown limitations in dealing with complex behavior patterns and changes in user interests, making it difficult to achieve high-precision personalized recommendations. In addition, the accumulation of user feedback data and the increase in real-time requirements have put forward higher requirements on the dynamic response capabilities of recommendation systems.

[0003] Therefore, there are still some key problems that need to be solved in traditional recommendation systems, which need to be improved through new algorithms and optimization technologies: First, traditional recommendation systems mainly rely on static similarity calculations between users and items, which makes it difficult to accurately capture the proportional differences and discreteness between features. As a result, when traditional recommendation systems deal with complex and diverse user behaviors, the accuracy of personalized recommendations is insufficient, resulting in recommendation results that are inconsistent with the user's real interests; secondly, traditional recommendation systems have weak responsiveness to changes in user interests, and most rely on single or short-term feedback to adjust strategies, which cannot effectively accumulate users' long-term interest preferences, easily leading to misjudgment of user needs by the recommendation system, and ignoring the gradual and dynamic development of user interests in multiple interactions; finally, traditional recommendation systems lack global performance optimization and regulation mechanisms, and most focus on the optimization of a single recommendation list, while ignoring the overall performance quantitative evaluation in multiple rounds of recommendations, causing the recommendation system to easily fall into local optimality and be unable to guarantee long-term performance and improvement of user satisfaction. Summary of the invention

[0004] The present invention provides a big data-driven intelligent recommendation system to solve the problems that traditional recommendation systems mainly rely on static similarity calculations between users and items, which makes it difficult to accurately capture the proportional differences and discreteness between features, resulting in insufficient accuracy of personalized recommendations when processing complex and diverse user behaviors, causing recommendation results to be inconsistent with the user's real interests; traditional recommendation systems have weak responsiveness to changes in user interests, mostly rely on single or short-term feedback for strategy adjustment, and are unable to effectively accumulate users' long-term interest preferences, which easily leads to the recommendation system misjudging user needs and ignoring the gradual changes and dynamic development of user interests in multiple interactions; traditional recommendation systems lack global performance optimization and regulation mechanisms, mostly focus on the optimization of a single recommendation list, and ignore the overall performance quantitative evaluation in multiple rounds of recommendations, resulting in the recommendation system easily falling into local optimality and being unable to guarantee long-term performance and improvement of user satisfaction.

[0005] The present invention provides a big data driven intelligent recommendation system, which specifically includes the following technical solutions:

[0006] A big data driven intelligent recommendation method comprises the following steps:

[0007] S1: Collect user behavior data and item feature data to generate user feature sets and item feature sets; based on the user feature sets and item feature sets, introduce user feature weights and calculate relative deviations; based on the relative deviations, calculate recommendation values ​​and generate recommendation lists; show the recommendation lists to users to obtain feedback values;

[0008] S2: Based on the feedback value, recommendation value and relative deviation, the cumulative feedback difference mechanism is adopted to dynamically adjust the user feature weights to obtain the final user feature weights; based on the final user feature weights, the relative deviation and recommendation value are calculated to finally generate a personalized recommendation list.

[0009] Preferably, the S1 specifically includes:

[0010] Based on the user feature set and the item feature set, the relative deviation is calculated according to the difference between the user features and the item features. The specific formula is as follows:

[0011]

[0012] Among them, D(F j , U i ) is the relative deviation, indicating the matching degree between the i-th user and the j-th item; F j is the item feature set of the jth item; U i is the user feature set of the i-th user; j is the item number; i is the user number; m is the number of user features; p is the index variable of the user features; w ipis the user feature weight; δ pq is the correlation coefficient between the p-th user feature and the q-th item feature; n is the number of item features; q is the index variable of the item feature; f jq represents the q-th item feature of the j-th item; u ip represents the p-th user feature of the i-th user; λ is the adjustment parameter; is the mean value of the user feature set U of the i-th user i obtained by averaging all the user feature values of the i-th user.

[0013] Preferably, the S1 specifically includes:

[0014] Based on the relative deviation degree, measure the matching degree between the item and the user's needs, and generate a recommendation value.

[0015] Preferably, the S1 specifically includes:

[0016] Calculate the recommendation values of all items, sort all items in descending order of the recommendation values, and generate a recommendation list; display the recommendation list to the user, monitor the user interaction behavior, and map the user interaction behavior to a quantitative feedback value.

[0017] Preferably, the S2 specifically includes:

[0018] The cumulative feedback difference mechanism, through iteration, accumulates the feedback differences of the user on the recommended items, dynamically adjusts the user feature weights, and obtains the updated user feature weights. The specific formula is as follows:

[0019]

[0020] where, w ip is the user feature weight, representing the updated user feature weight of the i-th user on the p-th feature; is the user feature weight of the i-th user on the p-th feature before update; δ pq is the correlation coefficient between the p-th user feature and the q-th item feature; ζ is the learning rate; P is the total number of iteration rounds that have been carried out; t is the index variable of the iteration round; is the feedback value of the i-th user on the j-th item in the t-th round; is the recommendation value of the j-th item in the t-th round.

[0021] Preferably, the S2 specifically includes:

[0022] Summarize the recommendation values of all recommended items, and introduce the relative deviation degree for weighted adjustment to obtain the total recommendation value of all recommended items.

[0023] Preferably, the S2 specifically includes:

[0024] Based on the total recommended value, the convergence condition is set. The specific formula is as follows:

[0025]

[0026] Where M is the total number of users participating in smart recommendation; N is the total number of items participating in smart recommendation; j is the item number; i is the user number; D(F j , U i ) is the relative deviation, indicating the matching degree between the i-th user and the j-th item; F j is the item feature set of the jth item; U i is the user feature set of the i-th user; θ is the global adjustment coefficient; is the total recommendation value of the tth iteration; is the total recommendation value of the t-1th iteration; ∈ is the convergence threshold.

[0027] Preferably, the S2 specifically includes:

[0028] After each round of iteration, check the convergence condition and the number of iterations. When the number of iterations is less than or equal to the preset maximum number of iterations and does not meet the convergence condition, use the updated user feature weights to recalculate the relative deviation and recommendation value, generate a new recommendation list, collect feedback values, and continue to use the cumulative feedback difference mechanism to dynamically adjust the user feature weights; when the number of iterations is greater than the preset maximum number of iterations or meets the convergence condition, stop the iteration, use the updated user feature weights as the final user feature weights, use the final user feature weights to calculate the relative deviation and recommendation value, and finally generate a personalized recommendation list. Preferably, S2 specifically includes:

[0029] Set a time interval to regularly collect user behavior data and item feature data, generate new user feature sets and item feature sets, and use the big data-driven intelligent recommendation method to recalculate the final user feature weights and generate a personalized recommendation list.

[0030] A big data driven intelligent recommendation system, including the following parts:

[0031] Data collection module, calculation and recommendation module, user feedback collection module, weight update module, personalized recommendation module;

[0032] The data collection module collects user behavior data and item feature data from the big data platform, generates and outputs user feature sets and item feature sets; the data collection module is connected to the calculation and recommendation module and the personalized recommendation module by means of data transmission;

[0033] The calculation and recommendation module receives the user feature set and the item feature set, introduces the user feature weight, and calculates the relative deviation; generates a recommendation value based on the relative deviation, and outputs the relative deviation and the recommendation value; sorts all items from high to low according to the recommendation value, generates a recommendation list and outputs it; the calculation and recommendation module is connected to the user feedback collection module and the weight update module by data transmission;

[0034] The user feedback collection module receives the recommendation list, displays the recommendation list to the user, and monitors the user interaction behavior; generates and outputs the feedback value according to the user interaction behavior; the user feedback collection module is connected to the weight update module by means of data transmission;

[0035] The weight update module receives feedback values, recommendation values ​​and relative deviations, and dynamically adjusts user feature weights based on the feedback values ​​and recommendation values ​​by using a cumulative feedback difference mechanism; calculates a total recommendation value based on the recommendation value and relative deviation; sets a convergence condition based on the total recommendation value; checks the convergence condition and the number of iterations after each round of iteration, and when the number of iterations is less than or equal to a preset maximum number of iterations and does not meet the convergence condition, inputs the updated user feature weights into the calculation and recommendation module, and continues to dynamically adjust the user feature weights; stops iteration when the number of iterations is greater than the preset maximum number of iterations or meets the convergence condition; inputs the updated user feature weights into the personalized recommendation module as final user feature weights; the weight update module is connected to the calculation and recommendation module and the personalized recommendation module by data transmission;

[0036] The personalized recommendation module receives the final user feature weight, the user feature set and the item feature set, calculates the relative deviation and the recommendation value based on the final user feature weight, and finally generates a personalized recommendation list and displays it to the user.

[0037] The beneficial effects of the technical solution of the present invention are:

[0038] 1. The present invention captures the proportional differences and discreteness of user features and item features in different dimensions by introducing the relative deviation formula, thereby achieving more accurate personalized recommendations. Compared with the traditional recommendation algorithm based on static similarity calculation between users and items, the present invention improves the sensitivity to the difference in feature proportions through normalization processing, so that the intelligent recommendation system can accurately reflect the user's interest in different features, avoiding the situation where the recommendation effect is poor due to the user's wide range of interests or discrete behavioral characteristics.

[0039] 2. The present invention adopts a cumulative feedback difference mechanism to dynamically adjust user feature weights; the cumulative feedback difference mechanism continuously accumulates user feedback information in multiple rounds of iterations, and uses the feedback differences to adjust user feature weights, so that the intelligent recommendation system can better adapt to changes in user immediate interests and long-term preferences; through multiple rounds of dynamic optimization, the deviation problem caused by over-reliance on single feedback is avoided, and the stability of the recommendation results and the ability to respond to changes in user needs are enhanced.

[0040] 3. The present invention realizes the global optimization of recommendation effect by introducing the convergence condition of total recommendation value calculation; the total recommendation value not only summarizes the recommendation effect of all recommended items, but also combines the weighted adjustment of relative deviation to provide a quantitative indicator for the overall performance of the intelligent recommendation system; the setting of convergence condition prevents invalid iteration and ensures the efficient operation of the intelligent recommendation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a structural diagram of a big data driven intelligent recommendation system according to the present invention;

[0042] Figure 2 This is a flow chart of a big data driven intelligent recommendation method described in the present invention. DETAILED DESCRIPTION

[0043] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0044] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0045] The following is a detailed description of a specific solution of a big data driven intelligent recommendation system provided by the present invention in conjunction with the accompanying drawings.

[0046] Refer to the attached Figure 1 , which shows a structure diagram of a big data driven intelligent recommendation system provided by an embodiment of the present invention, the system includes the following parts:

[0047] Data collection module, calculation and recommendation module, user feedback collection module, weight update module, personalized recommendation module;

[0048] The data collection module collects user behavior data and item feature data from the big data platform (such as user click stream logs, purchase records, and product databases), generates and outputs user feature sets and item feature sets; the data collection module is connected to the calculation and recommendation module and the personalized recommendation module through data transmission;

[0049] The calculation and recommendation module receives a user feature set and an item feature set, calculates a relative deviation using a preset initial value of a user feature weight or receives an updated user feature weight; generates a recommendation value based on the relative deviation, and outputs the relative deviation and the recommendation value; sorts all items from high to low according to the recommendation value, generates a recommendation list, and outputs it; the calculation and recommendation module is connected to the user feedback collection module and the weight update module by means of data transmission;

[0050] The user feedback collection module receives the recommendation list, displays the recommendation list to the user, and monitors the user interaction behavior; generates and outputs the feedback value according to the user interaction behavior; the user feedback collection module is connected to the weight update module by means of data transmission;

[0051] The weight update module receives feedback values, recommendation values ​​and relative deviations, and dynamically adjusts user feature weights based on the feedback values ​​and recommendation values ​​by using a cumulative feedback difference mechanism; calculates a total recommendation value based on the recommendation value and relative deviation; sets a convergence condition based on the total recommendation value; checks the convergence condition and the number of iterations after each round of iteration, and when the number of iterations is less than or equal to a preset maximum number of iterations and does not meet the convergence condition, inputs the updated user feature weights into the calculation and recommendation module, and continues to dynamically adjust the user feature weights; stops iteration when the number of iterations is greater than the preset maximum number of iterations or meets the convergence condition; inputs the updated user feature weights into the personalized recommendation module as final user feature weights; the weight update module is connected to the calculation and recommendation module and the personalized recommendation module by data transmission;

[0052] The personalized recommendation module receives the final user feature weight, the user feature set and the item feature set, calculates the relative deviation and the recommendation value based on the final user feature weight, and finally generates a personalized recommendation list and displays it to the user.

[0053] Refer to the attached Figure 2 , which shows a flow chart of a big data driven intelligent recommendation method provided by an embodiment of the present invention, the method comprising the following steps:

[0054] S1: Collect user behavior data and item feature data to generate user feature sets and item feature sets; based on the user feature sets and item feature sets, introduce user feature weights and calculate relative deviations; based on the relative deviations, calculate recommendation values ​​and generate recommendation lists; show the recommendation lists to users to obtain feedback values;

[0055] First, collect user behavior data from the big data platform (such as user click stream logs, purchase records, and product databases) as user features and form a user feature set. Collect item feature data as item features and form an item feature set.

[0056] The user features are used to describe the user's preferences, behaviors or attributes, and are extracted from the user's historical behaviors (such as clicks, browsing, purchases, etc.), for example: user interest categories, price preferences, purchase frequency, brand preferences. The user feature set is denoted as U i = {u i1 ,u i2 , ..., u im}, where m is the number of user features; i is the user ID; u ip represents the pth user feature of the ith user, where p ranges from 1 to m; the item feature is used to describe the static attributes of the pushed item and capture the specific information of the item, such as item category, price, brand, and sales volume. The item feature set is denoted as F j ={f j1 , f j2 , ..., f jn}, where n is the number of item features, j is the item number, and f jq represents the qth item feature of the jth item, where q ranges from 1 to n. The feature types and feature quantities of the user feature set and the item feature set are set according to the specific implementation scenario;

[0057] Then, based on the user feature set and the item feature set, the relative deviation is calculated according to the difference between the user features and the item features. The relative deviation is used to measure the matching degree between the user features and the item features, quantify the difference and proportional relationship between the user features and the item features in multiple dimensions, and enhance the sensitivity of the intelligent recommendation system to the proportional difference between different features through normalization processing, so as to more comprehensively capture the deviation between user interests and item attributes, so as to provide more refined feature matching.

[0058] The relative deviation formula is as follows:

[0059]

[0060] Among them, D(F j , U i ) is the relative deviation, indicating the matching degree between the i-th user and the j-th item; F j is the item feature set of the jth item; U iis the user feature set of the i-th user; j is the item number, which is the index variable of the pushed item; i is the user number, which is the index variable of the user; m is the number of user features, that is, the total number of user features included in the user feature set; p is the index variable of the user feature; w ip is the user feature weight, which indicates the preference of the i-th user for the p-th user feature. The initial value of the user feature weight is set according to the specific implementation scenario using the expert experience method; δ pq is the correlation coefficient between the pth user feature and the qth item feature, which is used to measure the correlation and influence of user features and item features, and is set according to the specific implementation scenario using expert experience; n is the number of item features, that is, the total number of item features included in the item feature set; q is the index variable of the item feature; f jq is the item feature, which represents the qth item feature of the jth item, expressed in scalar form; u ip is the user feature, which represents the pth user feature of the ith user, expressed in scalar form; λ is the adjustment parameter, which is used to control the influence of the mean square error term in the relative deviation, and is obtained by experimental tuning; is the user feature set U of the i-th user i The mean of is obtained by averaging all the user feature values ​​of the i-th user, reflecting the overall feature level of the i-th user; is the minimum value selection operation;

[0061] Next, a recommendation value of the item is generated according to the relative deviation, and the recommendation value is used to measure the degree of matching between the item and the user's needs;

[0062] The recommended value calculation formula is as follows:

[0063]

[0064] Among them, R j is the recommendation value of the jth item. The higher the recommendation value, the higher the matching degree between the item and the user. α is the scaling factor used to control the relative deviation D(F j , U i ) on the recommended value; β is a nonlinear coefficient, which is used to adjust the nonlinear change of the recommended value. The larger the nonlinear coefficient, the more sensitive the recommended value is to the change of relative deviation, that is, the increase of relative deviation will cause the recommended value to drop rapidly; the values ​​of scaling coefficient α and nonlinear coefficient β are obtained through experimental tuning;

[0065] Furthermore, after calculating the recommendation values ​​of all items, all items are sorted in descending order according to the recommendation values ​​to generate a recommendation list, so that items with the highest recommendation values ​​are displayed to users first. The recommendation list is displayed to users, and user interaction behaviors (such as clicks, browsing time, favorites, likes, purchases, etc.) are monitored. The user interaction behaviors are converted into quantitative feedback values ​​using a behavioral scoring model, and different types of feedback data are normalized to uniformly map the feedback values ​​to a fixed numerical range. The type of feedback data and the fixed numerical range can be set according to the specific implementation scenario. The behavioral scoring model is a prior art and will not be described in detail here.

[0066] S2: Based on the feedback value, recommendation value and relative deviation, the cumulative feedback difference mechanism is adopted to dynamically adjust the user feature weights to obtain the final user feature weights; based on the final user feature weights, the relative deviation and recommendation value are calculated to finally generate a personalized recommendation list.

[0067] Based on the feedback value and recommendation value, the cumulative feedback difference mechanism is adopted to dynamically adjust the user feature weights to obtain the updated user feature weights. The updated user feature weights will be used to recalculate the relative deviation and recommendation value in the next round of recommendations, so that the intelligent recommendation system can more accurately meet the needs of users.

[0068] The cumulative feedback difference mechanism accumulates the feedback differences of users on recommended items through multiple rounds of iterations, and dynamically adjusts the weights of user features, thereby capturing changes in users' immediate interests and long-term preferences, ensuring that the intelligent recommendation system can adapt to changes in user needs, avoiding deviations caused by over-reliance on single feedback, and improving the stability and responsiveness of the overall recommendation effect.

[0069] The weight update formula is as follows:

[0070]

[0071] Among them, w ip is the user feature weight, which represents the updated user feature weight of the i-th user on the p-th feature; is the user feature weight of the i-th user before updating the p-th feature; δ pq is the correlation coefficient between the pth user feature and the qth item feature; ζ is the learning rate, which is used to control the influence of user feedback on the update of user feature weights and is obtained through experimental tuning; T is the total number of iterations, indicating the number of iterations that have been performed; t is the index variable of the iteration; is the feedback value, which indicates the feedback value of the i-th user on the j-th item in the t-th round. The higher the feedback value, the more interested the user is. is the recommended value of the j-th item in round t.

[0072] The weight update formula dynamically adjusts the user feature weights by accumulating feedback differences. The changes in user feature weights are driven by user feedback in multiple rounds of interaction and regulated by the learning rate, which enhances the ability to capture user interests and improves the accuracy and personalization of recommendation results.

[0073] While dynamically adjusting the user feature weights through the cumulative feedback difference mechanism, the total recommendation value of all recommended items is calculated to measure the global performance of the recommendation effect and used to set subsequent convergence conditions. The total recommendation value summarizes the recommendation values ​​of all recommended items and combines the weighted adjustment of relative deviation to provide a quantitative basis for evaluating the global optimization effect.

[0074] The total recommended value is calculated as follows:

[0075]

[0076] Among them, R total is the total recommendation value of this iteration. The higher the total recommendation value, the better the intelligent recommendation system performs in satisfying user preferences and matching degree. j is the recommendation value of the jth item in this iteration; γ is the total recommendation value adjustment coefficient, which is used to control the sensitivity of the total recommendation value to the relative deviation and is obtained through experimental tuning; N is the total number of items participating in the intelligent recommendation and is set according to the specific implementation scenario.

[0077] The cumulative feedback difference mechanism continuously adjusts the user feature weights according to the feedback value in multiple rounds of iterations. In order to prevent falling into an invalid iteration cycle and improve efficiency and stability, the convergence condition of the cumulative feedback difference mechanism is set. The convergence condition formula is as follows:

[0078]

[0079] Where M is the total number of users participating in smart recommendation, which is set according to the specific implementation scenario; θ is the global adjustment coefficient, which is used to control the impact of changes in the total recommendation value on the convergence condition; is the total recommendation value of the tth iteration; is the total recommendation value of the t-1th iteration; ∈ is the convergence threshold, which is set according to the specific implementation scenario.

[0080] After each round of iteration, the convergence conditions and number of iterations are checked. When the number of iterations is less than or equal to the preset maximum number of iterations and the convergence conditions are not met, the updated user feature weights are used to recalculate the relative deviation and recommendation value, generate a new recommendation list, further collect feedback values, and continue to use the cumulative feedback difference mechanism to dynamically adjust the user feature weights.

[0081] When the number of iterations is greater than the preset maximum number of iterations or the convergence condition is met, the iteration is stopped and the updated user feature weight is used as the final user feature weight. The final user feature weight not only reflects the user's current interest tendency, but also captures the accumulation and change trend of the user's long-term preference in multiple rounds of interaction. In the subsequent real-time recommendation, the final user feature weight is used to calculate the relative deviation and recommendation value to generate a personalized recommendation list. The maximum number of iterations is set according to the specific implementation scenario.

[0082] Since the interests and behaviors of users will change dynamically, it is necessary to set a fixed time interval according to the specific implementation scenario, regularly collect the latest user behavior data and item feature data, and generate new user feature sets and item feature sets. Subsequently, the big data-driven intelligent recommendation method is used to recalculate the final user feature weights and generate a personalized recommendation list to ensure that the recommendation results are synchronized with the user's latest needs and improve the accuracy and personalization of the recommendation.

[0083] In summary, a big data-driven intelligent recommendation system has been completed.

[0084] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0085] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0086] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A big data driven intelligent recommendation method, characterized in that: The following steps are involved: S1: Collect user behavior data and item feature data to generate user feature sets and item feature sets; based on the user feature sets and item feature sets, introduce user feature weights and calculate relative deviations; based on the relative deviations, calculate recommendation values ​​and generate recommendation lists; Display the recommendation list to the user and obtain feedback value; S2: Based on the feedback value, recommendation value and relative deviation, the cumulative feedback difference mechanism is adopted to dynamically adjust the user feature weight to obtain the final user feature weight; Based on the feature weights of the end users, the relative deviation and recommendation value are calculated, and finally a personalized recommendation list is generated.

2. The big data driven intelligent recommendation method according to claim 1, characterized in that: The S1 specifically includes: Based on the user feature set and the item feature set, the relative deviation is calculated according to the difference between the user features and the item features. The specific formula is as follows: Among them, D(F j , U i ) is the relative deviation, indicating the matching degree between the i-th user and the j-th item; F j is the item feature set of the jth item; U i is the user feature set of the i-th user; j is the item number; i is the user number; m is the number of user features; p is the index variable of the user features; w ip is the user feature weight; δ pq is the correlation coefficient between the pth user feature and the qth item feature; n is the number of item features; q is the index variable of the item feature; f jq represents the qth item feature of the jth item; u ip represents the pth user feature of the i-th user; λ is the adjustment parameter; is the user feature set U of the i-th user i The mean of is obtained by averaging all user feature values ​​of the i-th user.

3. The big data driven intelligent recommendation method according to claim 2, characterized in that: The S1 specifically includes: Based on the relative deviation, the matching degree between items and user needs is measured to generate a recommendation value.

4. The big data driven intelligent recommendation method according to claim 3, characterized in that: The S1 specifically includes: Calculate the recommendation values ​​of all items, and sort all items in descending order according to the recommendation value to generate a recommendation list; display the recommendation list to the user, monitor user interaction behavior, and map user interaction behavior to quantitative feedback values.

5. The big data driven intelligent recommendation method according to claim 1, characterized in that: The S2 specifically includes: The cumulative feedback difference mechanism accumulates the feedback differences of users on recommended items through iteration, dynamically adjusts the user feature weights, and obtains the updated user feature weights. The specific formula is as follows: Among them, w ip is the user feature weight, which represents the updated user feature weight of the i-th user on the p-th feature; is the user feature weight of the i-th user before updating the p-th feature; δ pq is the correlation coefficient between the p-th user feature and the q-th item feature; ζ is the learning rate; T is the total number of iterations that have been performed; t is the index variable of the iteration; is the feedback value of the i-th user on the j-th item in the t-th round; is the recommended value of the j-th item in round t.

6. The big data driven intelligent recommendation method according to claim 5, characterized in that: The S2 specifically includes: Summarize the recommended values ​​of all recommended items, introduce relative deviation for weighted adjustment, and get the total recommended value of all recommended items.

7. The big data driven intelligent recommendation method according to claim 6, characterized in that: The S2 specifically includes: Based on the total recommended value, the convergence condition is set. The specific formula is as follows: Where M is the total number of users participating in smart recommendation; N is the total number of items participating in smart recommendation; j is the item number; i is the user number; D(F j , U i ) is the relative deviation, indicating the matching degree between the i-th user and the j-th item; F j is the item feature set of the jth item; U i is the user feature set of the i-th user; θ is the global adjustment coefficient; is the total recommendation value of the tth iteration; is the total recommendation value of the t-1th iteration; ∈ is the convergence threshold.

8. The big data driven intelligent recommendation method according to claim 7, characterized in that: The S2 specifically includes: After each round of iteration, the convergence conditions and number of iterations are checked. When the number of iterations is less than or equal to the preset maximum number of iterations and the convergence conditions are not met, the updated user feature weights are used to recalculate the relative deviation and recommendation value, generate a new recommendation list, collect feedback values, and continue to use the cumulative feedback difference mechanism to dynamically adjust the user feature weights. When the number of iterations is greater than the preset maximum number of iterations or the convergence conditions are met, the iteration is stopped, the updated user feature weights are used as the final user feature weights, the final user feature weights are used to calculate the relative deviation and recommendation value, and finally a personalized recommendation list is generated.

9. The big data driven intelligent recommendation method according to claim 8, characterized in that: The S2 specifically includes: Set a time interval to regularly collect user behavior data and item feature data, generate new user feature sets and item feature sets, and use the big data-driven intelligent recommendation method to recalculate the final user feature weights and generate a personalized recommendation list.

10. A big data driven intelligent recommendation system, applied to the big data driven intelligent recommendation method as claimed in claim 1, characterized in that: Includes the following sections: Data collection module, calculation and recommendation module, user feedback collection module, weight update module, personalized recommendation module; The data collection module collects user behavior data and item feature data from the big data platform, generates user feature sets and item feature sets, and outputs them; The data collection module is connected with the calculation and recommendation module and the personalized recommendation module through data transmission; The calculation and recommendation module receives the user feature set and the item feature set, introduces the user feature weight, and calculates the relative deviation; Based on the relative deviation, a recommended value is generated, and the relative deviation and the recommended value are output; Sort all items by recommendation value from high to low, generate a recommendation list and output it; The calculation and recommendation module is connected with the user feedback collection module and the weight update module through data transmission; The user feedback collection module receives the recommendation list, displays the recommendation list to the user, and monitors the user's interactive behavior; Generate feedback values ​​and output them based on user interaction behaviors; The user feedback collection module is connected to the weight update module through data transmission; The weight updating module receives the feedback value, the recommendation value and the relative deviation, and dynamically adjusts the user feature weight based on the feedback value and the recommendation value by adopting the cumulative feedback difference mechanism; and calculates the total recommendation value based on the recommendation value and the relative deviation; Based on the total recommended value, set the convergence condition; After each round of iteration, the convergence condition and the number of iterations are checked. When the number of iterations is less than or equal to the preset maximum number of iterations and the convergence condition is not met, the updated user feature weights are input into the calculation and recommendation module to continue to dynamically adjust the user feature weights. When the number of iterations is greater than the preset maximum number of iterations or the convergence condition is met, the iteration is stopped; the updated user feature weight is input into the personalized recommendation module as the final user feature weight; The weight update module is connected with the calculation and recommendation module and the personalized recommendation module through data transmission; The personalized recommendation module receives the final user feature weight, the user feature set and the item feature set, calculates the relative deviation and the recommendation value based on the final user feature weight, and finally generates a personalized recommendation list and displays it to the user.

Images