Intelligent push method and system based on dynamic interest decay analysis

Abstract

The application provides an intelligent pushing method and system based on dynamic interest decay analysis, relates to the technical field of data pushing, and comprises the following steps: when a target user triggers a video recommendation request, determining a first recommended video list based on the video recommendation request and historical viewing records of the target user; obtaining first list viewing features of the target user for the first recommended video list, performing dynamic interest decay analysis based on the first list viewing features, and determining a second recommended video list according to an interest decay analysis result; and recommending the second recommended video list to a user terminal of the target user. The technical problem that the short video recommendation content deviates from the real-time interest of a user in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data pushing, in particular to an intelligent pushing method and system based on dynamic interest decay analysis. BACKGROUND

[0002] In the process of continuous watching of short videos by a user, the interest in a specific category of content tends to naturally decay over time. However, traditional short video recommendation algorithms mostly rely on historical behavior data of the user to construct a static recommendation model, which is difficult to capture the dynamic evolution of the user's interest in real time, and thus lacks effective perception and response to the interest decay trend, thereby often leading to deviation of the recommended content from the real-time interest of the user. SUMMARY

[0003] The present application provides an intelligent pushing method and system based on dynamic interest decay analysis to solve the technical problem of deviation of the recommended content of short videos from the real-time interest of the user in the prior art.

[0004] The technical solution of the present application to solve the above technical problem is as follows:

[0005] In a first aspect, the present application provides an intelligent pushing method based on dynamic interest decay analysis, comprising:

[0006] determining a first recommended video list based on a video recommendation request and historical viewing records of a target user when the target user triggers the video recommendation request;

[0007] obtaining first list viewing features of the target user for the first recommended video list, performing dynamic interest decay analysis based on the first list viewing features, and determining a second recommended video list according to the interest decay analysis result;

[0008] recommending the second recommended video list to a user terminal of the target user.

[0009] In a second aspect, the present application provides an intelligent pushing system based on dynamic interest decay analysis, comprising:

[0010] a first recommendation module configured to determine a first recommended video list based on a video recommendation request and historical viewing records of a target user when the target user triggers the video recommendation request;

[0011] an interest decay analysis module configured to obtain first list viewing features of the target user for the first recommended video list, perform dynamic interest decay analysis based on the first list viewing features, and determine a second recommended video list according to the interest decay analysis result;

[0012] a second recommendation module configured to recommend the second recommended video list to a user terminal of the target user.

[0013] The beneficial effects of the present application are:

[0014] Compared with the prior art, first, when the target user triggers a video recommendation request, a first recommended video list is determined based on the video recommendation request and the historical viewing records of the target user, which conforms to the historical preferences of the target user and provides a reliable basis for subsequent dynamic interest decay analysis through real-time feedback. Secondly, the first list viewing features of the target user for the first recommended video list are obtained, dynamic interest decay analysis is performed based on the first list viewing features, and a second recommended video list is determined according to the interest decay analysis result. The current interest of the target user is quantified by the ratio of the current viewing time length average to the historical viewing time length average, and the interest weight is updated accordingly to generate the second recommended video list, which realizes the change of the recommendation strategy from static historical driving to dynamic real-time feedback driving, and ensures that the recommended content is always synchronized with the current interest of the user. Finally, the second recommended video list is recommended to the user end of the target user, ensuring that the recommended video received by the user is the result of real-time optimization and current interest, improving the acceptance of the recommended content and the viewing experience of the user.

[0015] Through the above technical solution, the first recommended video list is generated according to the historical viewing records of the target user, and the real-time viewing features of the target user for the first recommended video list are recorded, the difference between the current interest and the historical interest is analyzed, the interest decay or enhancement trend is quantified, and the interest weight is adjusted to generate a second recommended video list that is more in line with real-time preferences and is pushed. In this way, the recommended content is accurately matched with the current interest of the user, the recommendation deviation is reduced, and the viewing experience of the user is improved. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The flowchart of the intelligent push method based on dynamic interest decay analysis provided by the present application is shown in the figure.

[0017] Figure 2 The structure diagram of the intelligent push system based on dynamic interest decay analysis provided by the present application is shown in the figure.

[0018] In the drawings, the components represented by each reference numeral are as follows:

[0019] The first recommendation module 11, the interest decay analysis module 12, and the second recommendation module 13. DETAILED DESCRIPTION

[0020] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative effort are within the scope of the present application.

[0021] In the description of the present application, the terms "first", "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise specifically limited.

[0022] In the description of the present application, the term "for example" is used to indicate "as an example, illustration or description". Any embodiment described as "for example" in the present application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given so that any person skilled in the art can implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can realize the present application without using these specific details. In other examples, well-known structures and processes will not be described in detail to avoid unnecessary details making the description of the present application obscure. Therefore, the present application is not intended to be limited to the shown embodiments, but is consistent with the broadest scope in accordance with the principles and characteristics disclosed.

[0023] Embodiment one, as shown in the present application, provides an intelligent pushing method based on dynamic interest decay analysis, comprising: Figure 1

[0024] S10: When a target user triggers a video recommendation request, a first recommended video list is determined based on the video recommendation request and the historical viewing records of the target user.

[0025] The historical viewing records of the target user often contain his past interest preferences, including historical preference features such as preferred content categories and focused theme directions. By extracting and analyzing these historical viewing records, the interest categories and corresponding interest weights of the target user can be identified, and a first recommended video list that meets the past interest features of the target user can be generated accordingly.

[0026] ​To solve the above problems, the video recommendation request is triggered when the target user clicks any video in the short video application to enter the video playing interface. In this way, the video recommendation request is bound with the active interaction of the user, ensuring that the starting of the recommendation process is synchronized with the actual viewing behavior of the user. Further, when the target user triggers the video recommendation request, a first recommended video list is determined based on the video recommendation request and the historical viewing records of the target user.

[0027] Specifically, step S10 in the method comprises:

[0028] The full video pool is connected according to the video recommendation request, and the full video pool contains all the recommendable videos;

[0029] The historical viewing features are extracted according to the historical viewing records of the target user, and a plurality of user interest categories and interest weights of each user interest category are determined based on the historical viewing features;

[0030] The video selection is performed in the full video pool based on the plurality of user interest categories and the interest weights of each user interest category, and the first recommended video list is obtained.

[0031] In the embodiment of the application, the full video pool is first connected according to the video recommendation request, and the full video pool contains all the recommendable videos of the short video application platform, ensuring that the source of the recommended content is comprehensive.

[0032] Secondly, the historical viewing features are extracted according to the historical viewing records of the target user, and a plurality of user interest categories and interest weights of each user interest category are determined based on the historical viewing features, wherein the interest category refers to a video category such as food, travel, and fitness, and the interest weight of each user interest category is calculated. The interest weight can reflect the preference degree of the target user for the interest category, and the higher the interest weight, the stronger the preference of the target user for the interest category in the past.

[0033] Finally, the video selection is performed in the full video pool based on the plurality of user interest categories and the interest weights of each user interest category, and the first recommended video list is obtained. The first recommended video list is an initial recommendation result generated based on the historical preference of the target user, and provides basic data for subsequent dynamic interest decay analysis.

[0034] Specifically, the "historical viewing features are extracted according to the historical viewing records of the target user, and a plurality of user interest categories and interest weights of each user interest category are determined based on the historical viewing features" comprises:

[0035] The first historical viewing record is extracted by traversing the historical viewing records.

[0036] when the viewing duration of the first historical viewing record exceeds a preset duration threshold, obtaining a first video category of the first historical viewing record;

[0037] incrementing, according to the first video category, an occurrence frequency of the first video category in a full video category;

[0038] after processing the historical viewing records, taking a video category with a non-zero occurrence frequency in the full video category as a plurality of user interest categories of the target user;

[0039] calculating a ratio of the occurrence frequency of each user interest category to a sum of occurrence frequencies of all user interest categories as an interest weight of the corresponding user interest category.

[0040] In the embodiments of the application, the historical viewing records are first traversed to extract a first historical viewing record. For example, 100 historical viewing records of a target user are traversed, and detailed information of each historical viewing record is read, including a video ID, a viewing duration, and the like.

[0041] Secondly, when the viewing duration of the first historical viewing record exceeds a preset duration threshold, a first video category of the first historical viewing record is obtained, wherein the preset duration threshold can be flexibly set according to an actual scene, such as 10 seconds, 15 seconds, and the like, and is mainly used to filter invalid viewing behaviors such as mis-click and second slip, and the video category can be determined based on a classification system preset by a short video application platform, such as food, travel, fitness, music, and cute pets. For example, if the preset duration threshold is 10 seconds, 100 historical viewing records of a target user are traversed, 60 records with a viewing duration exceeding 10 seconds are selected as effective historical viewing records, and a video category corresponding to the effective historical viewing records is extracted, for example, 30 records correspond to food, 20 records correspond to travel, and 10 records correspond to fitness, so that food, travel, and fitness are taken as the first video category.

[0042] Thirdly, according to the first video category, an occurrence frequency of the first video category in a full video category is incremented. For example, if the frequencies of food, travel, and fitness in the full video category are 12 times, 8 times, and 3 times respectively, according to 60 effective historical viewing records (30 food, 20 travel, and 10 fitness), the occurrence frequency of the first video category is accumulated, and the frequency of food is 12+30=42 times, the frequency of travel is 8+20=28 times, and the frequency of fitness is 3+10=13 times.

[0043] Further, after processing the historical viewing records, the video categories with a frequency of occurrence of 0 in the full video category are determined as the multiple user interest categories of the target user. For example, after the above processing, the frequencies of the food, travel, and fitness categories are 42, 28, and 13, respectively, and the three categories are determined as the interest categories of the target user.

[0044] Finally, the ratio of the frequency of occurrence of each user interest category to the sum of the frequencies of occurrence of all user interest categories is calculated as the interest weight of the corresponding user interest category. For example, the sum of the frequencies of occurrence of the food, travel, and fitness categories is 42+28+13=83, and the interest weight of the food category is 42 / 83≈0.506, the interest weight of the travel category is 28 / 83≈0.337, and the interest weight of the fitness category is 13 / 83≈0.157. In this way, the invalid behavior is filtered by the preset time threshold, and the preference degree of the user for different categories is quantified by the frequency ratio, ensuring that the interest categories and weights can truly reflect the historical stable preferences of the user and avoiding the interference of accidental behavior on the recommendation logic.

[0045] Further, the "selecting videos in the full video pool based on the multiple user interest categories and the interest weights of the user interest categories to obtain the first recommended video list" comprises:

[0046] obtaining a preset recommended video number;

[0047] obtaining a preset recommended video number;

[0048] randomly selecting videos in the corresponding video category of the full video pool based on the preset recommended number of each user interest category to obtain the first recommended video list.

[0049] In the embodiment of the application, a preset recommended video number is first obtained, wherein the preset recommended video number is the total number of recommended videos per batch that the short video application platform sets in advance according to historical browsing data and other factors, such as 20, to control the length of the recommended list and avoid the user's selection burden due to too many videos or insufficient content supply due to too few videos. For example, if the short video application platform sets 20 videos per batch, the preset recommended video number is 20.

[0050] Secondly, according to the preset recommendation list video number and the interest weight of each user interest category, the preset recommendation number of each user interest category is obtained, wherein the preset recommendation number = preset recommendation list video number * interest weight, and the calculation result is fine-tuned in the ways of rounding off, rounding up, rounding down and the like to ensure that the sum of the preset recommendation numbers of each user interest category is equal to the preset recommendation list video number. Exemplarily, if the preset recommendation list video number is 20, the interest weight of food is 0.506, the interest weight of travel is 0.337, and the interest weight of fitness is 0.157, then the preset recommendation number of food = 20 * 0.506 ≈ 10 (rounding off), the preset recommendation number of travel = 20 * 0.337 ≈ 7 (rounding off), and the preset recommendation number of fitness = 20 * 0.157 ≈ 3 (rounding off). The sum of the preset recommendation numbers of food, travel and fitness is 20, which is consistent with the preset recommendation list video number. In this way, the interest weight is converted into a specific preset recommendation number to ensure that the proportion of the recommended content matches the user's historical preferences.

[0051] Finally, based on the preset recommendation number of each user interest category, a video is randomly selected in the corresponding video category in the full video pool to obtain a first recommended video list. Specifically, after determining the preset recommendation number of each interest category, a video is randomly selected from the video subcategory of the corresponding video category in the full video pool to form the first recommended video list. The purpose of random selection is to avoid homogenization of recommended content and ensure that the user sees more diverse content. Exemplarily, in the full video pool, the food category contains 1000 videos, the travel category contains 800 videos, and the fitness category contains 500 videos. According to the preset recommendation number, 10 videos are randomly selected from the video subcategory of food, 7 videos are randomly selected from the video subcategory of travel, and 3 videos are randomly selected from the video subcategory of fitness. These 20 videos are randomly arranged or alternately arranged according to the interest category to form the first recommended video list. The first recommended video list is an initial recommendation result generated based on the historical preferences of the target user.

[0052] In summary, compared with the prior art, when the target user triggers a video recommendation request, the first recommended video list is determined based on the video recommendation request and the historical viewing records of the target user. In this way, the interest features are extracted based on historical effective behaviors, and the first recommended video list that meets the historical preferences of the target user is finally generated, providing a reliable basis for subsequent dynamic interest decay analysis through real-time feedback.

[0053] S20: Obtain the first list viewing features of the target user on the first recommended video list, perform dynamic interest decay analysis based on the first list viewing features, and determine a second recommended video list according to the interest decay analysis result.

[0054] The interest of a user can dynamically change with content exposure, viewing duration, etc. during the process of watching a video. Traditional recommendation methods mostly rely on historical behaviors of a user to build a static recommendation model, which is difficult to capture the dynamic change in real time, and often continuously push content that the user has already lost interest in, resulting in a disconnection between the recommended content and the current real interest of the user.

[0055] To solve the above problems, the present application obtains first list viewing features of the target user for the first recommended video list, performs dynamic interest decay analysis based on the first list viewing features, and determines a second recommended video list according to the interest decay analysis result.

[0056] Specifically, step S20 in the method comprises:

[0057] Continuously record the viewing duration of the target user for each video in the first recommended video list to obtain the first list viewing features;

[0058] Perform user interest analysis on the first list viewing features to obtain the current interest features of the user;

[0059] Retrieve historical interest features of the user, combine the current interest features of the user to perform dynamic interest decay analysis, and determine a second recommended video list according to the interest decay analysis result.

[0060] In the embodiment of the present application, first, the viewing duration of the target user for each video in the first recommended video list is continuously recorded to obtain the first list viewing features. For example, when the target user watches the videos in the first recommended video list, the viewing duration of each video is recorded in real time and continuously, for example, the target user spends 25 seconds on the first food video and 40 seconds on the second travel video, etc. in the first recommended video list, and these scattered duration data are sequentially aggregated, such as {25 seconds, 40 seconds, …}, to obtain the first list viewing features, which can reflect the real-time interest of the target user during the watching process.

[0061] Secondly, the first list viewing features are analyzed for user interest to obtain the current interest features of the user. Specifically, the viewing duration average of all videos in each user interest category in the first recommended video list is calculated to obtain the viewing duration average of multiple user interest categories as the current interest features of the user, which can reflect the current preference degree of the user for each interest category.

[0062] Finally, the user historical interest features are invoked, the dynamic interest decay analysis is performed in combination with the current user interest features, and the second recommended video list is determined according to the interest decay analysis result. Specifically, the ratio of the average viewing time of the current user interest category to the average historical viewing time is calculated to obtain the interest adjustment coefficient of the current user interest category, the corresponding interest weight is adjusted according to the interest adjustment coefficient, the adjusted interest weight of each user interest category is obtained as the interest decay analysis result, and the second recommended video list is determined according to the interest decay analysis result.

[0063] Specifically, the "user interest analysis on the first list viewing features to obtain the current user interest features" includes:

[0064] The first user interest category is obtained from the plurality of user interest categories.

[0065] The average viewing time of all videos belonging to the first user interest category in the first recommended video list is calculated to obtain the first category average viewing time.

[0066] The average viewing time of the remaining user interest categories is obtained in the same way as the first category average viewing time, and the average viewing time of the plurality of user interest categories is obtained as the current user interest features.

[0067] In the embodiment of the application, the first user interest category is obtained from the plurality of user interest categories. For example, a random interest category, such as food, is selected from the plurality of user interest categories of the target user as the first user interest category.

[0068] Secondly, the average viewing time of all videos belonging to the first user interest category in the first recommended video list is calculated to obtain the first category average viewing time. For example, the viewing time of all videos belonging to the first user interest category in the first recommended video list is counted, such as the viewing time of 10 food videos in the first recommended video list of the target user is 25 seconds, 30 seconds, 20 seconds, …, the total viewing time is 250 seconds, and the average viewing time is 25 seconds, which is the first category average viewing time.

[0069] Finally, the average viewing time of the remaining user interest categories is obtained in the same way as the average viewing time of the first category, and the average viewing time of the multiple user interest categories is obtained as the current interest feature of the user. For example, the average viewing time of the remaining user interest categories is calculated in the same way, for example, the average viewing time of 7 videos in the travel category is 40 seconds, and the average viewing time of 3 videos in the fitness category is 20 seconds, and the user's current interest feature is composed of the average viewing time of food 25 seconds, the average viewing time of travel 40 seconds, and the average viewing time of fitness 20 seconds. In this way, the current interest is quantified by the average viewing time, and the longer the average viewing time, the higher the user's current interest in the category; otherwise, the interest may decrease.

[0070] Further, the "retrieving the user historical interest feature, combining the user current interest feature to perform dynamic interest decay analysis, and determining the second recommended video list according to the interest decay analysis result" comprises:

[0071] retrieving the user historical interest feature, the user historical interest feature comprising the historical average viewing time of each user interest category;

[0072] traversing the user current interest feature to obtain the average viewing time of the current user interest category, and obtaining the historical average viewing time of the current user interest category in the user historical interest feature;

[0073] calculating the ratio of the average viewing time of the current user interest category to the historical average viewing time to obtain the interest adjustment coefficient of the current user interest category;

[0074] obtaining the interest adjustment coefficient of each user interest category according to the way of obtaining the interest adjustment coefficient of the current user interest category;

[0075] adjusting the interest weight of each user interest category based on the interest adjustment coefficient of each user interest category to obtain the adjusted interest weight of each user interest category as the interest decay analysis result;

[0076] determining the second recommended video list according to the adjusted interest weight of each user interest category.

[0077] In the embodiments of the present application, first, the user historical interest features are invoked, and the user historical interest features include the historical viewing time average of each user interest category. For example, the historical viewing time average of each user interest category of the target user in the past period, such as the last three days, one week, one month, etc. is extracted, for example, the historical viewing time average of the target user's food in the last week is 30 seconds, the historical viewing time average of travel is 25 seconds, and the historical viewing time average of fitness is 25 seconds, which are used as the user historical interest features. The user historical interest features can reflect the stable preference state of the target user for different interest categories in the historical stage.

[0078] Secondly, the user current interest features are traversed to obtain the viewing time average of the current user interest category, and the historical viewing time average of the current user interest category is obtained from the user historical interest features. For example, the viewing time average of the current user interest category is traversed, for example, the viewing time average of food is 25 seconds, the viewing time average of travel is 40 seconds, and the viewing time average of fitness is 20 seconds, and the corresponding historical viewing time average of food is 30 seconds, the historical viewing time average of travel is 25 seconds, and the historical viewing time average of fitness is 25 seconds, which are obtained from the user historical interest features.

[0079] Thirdly, the ratio of the viewing time average of the current user interest category to the historical viewing time average is calculated to obtain the interest adjustment coefficient of the current user interest category. For example, the current interest adjustment coefficient of food is 25 / 30=0.83, the current interest adjustment coefficient of travel is 40 / 25=1.6, and the current interest adjustment coefficient of fitness is 20 / 25=0.8. If the interest adjustment coefficient>1, it indicates that the current interest in the interest category is increasing, if the interest adjustment coefficient≈1, it indicates that the current interest in the interest category is basically stable, and if the interest adjustment coefficient<1, it indicates that the current interest in the interest category is decaying.

[0080] Further, the interest weight of each user interest category is adjusted based on the interest adjustment coefficient of each user interest category to obtain the adjusted interest weight of each user interest category as the interest decay analysis result. For example, if the interest weights of each user interest category are food interest weight 0.506, travel interest weight 0.337, and fitness interest weight 0.157, then the adjusted interest weight of food is 0.506*0.83=0.42, the adjusted interest weight of travel is 0.337*1.6=0.54, and the adjusted interest weight of fitness is 0.157*0.8=0.13. As the interest decay analysis result, the interest decay analysis result can reflect the real-time state and evolution trend of the target user's interest in the dynamic change.

[0081] Finally, the second recommended video list is determined according to the adjusted interest weight of each user interest category. Exemplarily, the product of the preset recommended list video number and the adjusted interest weight of each user interest category is calculated in the same way as S10, and the calculation result is fine-tuned in the manner of rounding off, rounding up, rounding down, etc. to obtain the recommended number of each user interest category, and then the videos are randomly selected in the corresponding video category in the full video pool based on the recommended number of each user interest category to obtain the second recommended video list. For example, if the preset recommended list video number is 20, the recommended number of food = 20 x 0.42 = 8 (rounded down), the recommended number of travel = 20 x 0.54 = 10 (rounded down), and the recommended number of fitness = 20 x 0.13 = 2 (rounded down), and then the videos are randomly selected in the corresponding video category in the full video pool, for example, 8 videos are randomly selected from the food video subset, 10 videos are randomly selected from the travel video subset, and 2 videos are randomly selected from the fitness video subset, and the 20 videos are randomly arranged or alternately arranged according to the interest category to form the second recommended video list. The second recommended video list is a recommended result generated based on the current preference of the target user. In this way, the video recommendation strategy can be dynamically adjusted according to the real-time state of the target user's interest, the recommended number of the category with increased interest is increased, and the recommended number of the category with decreased interest is decreased.

[0082] In summary, compared with the prior art, the present application obtains the first list viewing feature of the target user on the first recommended video list, performs dynamic interest decay analysis based on the first list viewing feature, and determines the second recommended video list according to the interest decay analysis result. In this way, the current interest of the target user is quantified by the ratio of the current viewing time average to the historical viewing time average, the interest weight is updated accordingly, and the second recommended video list is generated, realizing the transition of the recommendation strategy from static historical driving to dynamic real-time feedback driving, and ensuring that the recommended content is always synchronized with the current interest of the user.

[0083] S30: recommending the second recommended video list to the user end of the target user.

[0084] The foregoing steps generate the second recommended video list through dynamic interest decay analysis. The second recommended video list is an optimized result obtained by dynamically adjusting based on the real-time viewing behavior of the user on the first recommended video list and the historical interest feature, and can more accurately reflect the current interest preference of the target user than the initial recommendation.

[0085] Therefore, in the embodiment of the present application, the second recommended video list is recommended to the user end of the target user, wherein the user end specifically refers to the terminal carrier actually used by the target user, such as the short video APP interface or the video playback page of the web end on the mobile phone, tablet and other devices.

[0086] In this way, the recommended content received by the user is ensured to be optimized in real time and highly matched with the current preference, thereby improving the acceptance of the recommended content and the overall viewing experience.

[0087] To sum up, the embodiments of the present application have at least the following technical effects:

[0088] Compared with the prior art, the present application first determines a first recommended video list based on the video recommendation request and the historical viewing records of the target user when the target user triggers a video recommendation request. In this way, the interest features are extracted based on the historical effective behaviors, and finally a first recommended video list that meets the historical preferences of the target user is generated, providing a reliable basis for subsequent dynamic interest decay analysis through real-time feedback.

[0089] Secondly, the present application obtains the first list viewing features of the target user on the first recommended video list, performs dynamic interest decay analysis based on the first list viewing features, and determines a second recommended video list according to the interest decay analysis result. In this way, the current interest of the target user is quantified by the ratio of the current viewing time average to the historical viewing time average, and the interest weight is updated accordingly to generate the second recommended video list, realizing the transition of the recommendation strategy from static historical driving to dynamic real-time feedback driving, and ensuring that the recommended content is always synchronized with the current interest of the user.

[0090] Finally, the present application recommends the second recommended video list to the user end of the target user. In this way, it is ensured that the recommended video received by the user is a result optimized in real time and in line with the current interest, improving the acceptance of the recommended content and the viewing experience of the user.

[0091] Through the above technical solution, the present application first generates a first recommended video list according to the historical viewing records of the target user, then records the real-time viewing features of the target user on the first recommended video list, analyzes the difference between the current interest and the historical interest, quantifies the interest decay or enhancement trend, and then adjusts the interest weight to generate a second recommended video list that is more in line with the real-time preference and pushes it. In this way, the recommended content is accurately matched with the current interest of the user, the recommendation deviation is reduced, and the viewing experience of the user is improved.

[0092] Embodiment two, as Figure 2 shown, based on the same inventive concept of the intelligent pushing method based on dynamic interest decay analysis provided in embodiment one, the present embodiment further provides an intelligent pushing system based on dynamic interest decay analysis, comprising:

[0093] a first recommendation module 11 for determining a first recommended video list based on the video recommendation request and the historical viewing records of the target user when the target user triggers a video recommendation request;

[0094] The interest decay analysis module 12 is configured to acquire first list viewing features of the target user on the first recommended video list, perform dynamic interest decay analysis based on the first list viewing features, and determine a second recommended video list according to the interest decay analysis result.

[0095] The second recommendation module 13 is configured to recommend the second recommended video list to a user terminal of the target user.

[0096] The first recommendation module 11 is specifically configured to:

[0097] The video recommendation request is triggered when the target user clicks any video in a short video application to enter a video playing interface.

[0098] Specifically, the "determining a first recommended video list based on the video recommendation request and the historical viewing records of the target user" includes:

[0099] According to the video recommendation request, a full-amount video pool is connected, and the full-amount video pool contains all the recommendable videos.

[0100] According to the historical viewing records of the target user, historical viewing features are extracted, and a plurality of user interest categories and interest weights of each user interest category are determined based on the historical viewing features.

[0101] Based on the plurality of user interest categories and the interest weights of each user interest category, video selection is performed in the full-amount video pool to obtain the first recommended video list.

[0102] Further, the "extracting historical viewing features from the historical viewing records of the target user, and determining a plurality of user interest categories and interest weights of each user interest category based on the historical viewing features" includes:

[0103] The first historical viewing record is extracted by traversing the historical viewing records.

[0104] When the viewing duration of the first historical viewing record exceeds a preset duration threshold, a first video category of the first historical viewing record is acquired.

[0105] The occurrence frequency of the first video category in the full-amount video category is incremented by 1 according to the first video category.

[0106] After processing the historical viewing records, the video categories with a non-zero occurrence frequency in the full-amount video category are taken as the plurality of user interest categories of the target user.

[0107] The ratio of the occurrence frequency of each user interest category to the total sum of the occurrence frequencies of all user interest categories is calculated as the interest weight of the corresponding user interest category.

[0108] Further, the "selecting videos in the full video pool based on the plurality of user interest categories and the interest weights of the user interest categories to obtain the first recommended video list" comprises:

[0109] obtaining a preset number of recommended list videos;

[0110] obtaining a preset recommended number of each of the user interest categories according to the preset number of recommended list videos and the interest weight of each of the user interest categories;

[0111] randomly selecting videos in the corresponding video category of the full video pool based on the preset recommended number of each of the user interest categories to obtain the first recommended video list.

[0112] The interest decay analysis module 12 is specifically configured to:

[0113] continuously record the viewing duration of each video in the first recommended video list watched by the target user to obtain first list viewing features;

[0114] performing user interest analysis on the first list viewing features to obtain user current interest features;

[0115] retrieving user historical interest features, combining the user current interest features to perform dynamic interest decay analysis, and determining a second recommended video list according to the interest decay analysis result.

[0116] Specifically, the "performing user interest analysis on the first list viewing features to obtain user current interest features" comprises:

[0117] obtaining a first user interest category from the plurality of user interest categories;

[0118] calculating the viewing duration average of all videos in the first recommended video list belonging to the first user interest category to obtain a first category viewing duration average;

[0119] obtaining the viewing duration average of the remaining user interest categories in the same way as obtaining the first category viewing duration average to obtain the viewing duration average of the plurality of user interest categories as the user current interest features.

[0120] Further, the "retrieving user historical interest features, combining the user current interest features to perform dynamic interest decay analysis, and determining a second recommended video list according to the interest decay analysis result" comprises:

[0121] retrieving user historical interest features, the user historical interest features comprising historical viewing duration averages of each of the user interest categories;

[0122] traversing the current interest characteristics of the user, obtaining a viewing time average of a current user interest category, and obtaining a historical viewing time average of the current user interest category in the historical interest characteristics of the user;

[0123] calculating a ratio of the viewing time average of the current user interest category to the historical viewing time average, to obtain an interest adjustment coefficient of the current user interest category;

[0124] obtaining the interest adjustment coefficient of each user interest category according to the manner of obtaining the interest adjustment coefficient of the current user interest category;

[0125] adjusting the interest weight of each user interest category based on the interest adjustment coefficient of each user interest category to obtain an adjusted interest weight of each user interest category as the interest decay analysis result;

[0126] determining a second recommended video list according to the adjusted interest weight of each user interest category.

[0127] The second recommendation module 13 is specifically configured to:

[0128] recommend the second recommended video list to the user terminal of the target user.

[0129] In summary, the embodiments of the present application have at least the following technical effects:

[0130] Compared with the prior art, the present application first determines a first recommended video list based on the video recommendation request and the historical viewing record of the target user through the first recommendation module when the target user triggers a video recommendation request. The first recommended video list meets the historical preferences of the target user, providing a reliable basis for subsequent dynamic interest decay analysis through real-time feedback. Secondly, through the interest decay analysis module, the first list viewing characteristics of the target user for the first recommended video list are obtained, dynamic interest decay analysis is performed based on the first list viewing characteristics, and a second recommended video list is determined according to the interest decay analysis result. The current interest of the target user is quantified by the ratio of the current viewing time average to the historical viewing time average, the interest weight is updated accordingly, and the second recommended video list is generated. This realizes the transition of the recommendation strategy from static historical driving to dynamic real-time feedback driving, ensuring that the recommended content is always synchronized with the current interest of the user. Finally, through the second recommendation module, the second recommended video list is recommended to the user terminal of the target user, ensuring that the recommended video received by the user is the result of real-time optimization and adaptation to the current interest, improving the acceptance of the recommended content by the user and the viewing experience. In this way, the recommended content is accurately matched with the current interest of the user, the recommendation deviation is reduced, and the viewing experience of the user is improved.

[0131] It should be noted that the above-mentioned embodiments have been described by way of example only, and that modifications and additions can be made thereto without departing from the scope of the application.

[0132] Those skilled in the art will appreciate that embodiments of the application can be situated as methods, systems or computer program products. Accordingly, the application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer-readable program code.

[0133] The application is described with reference to the flowchart and / or block diagram illustrations of the methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagrams, and combinations of blocks in the flowchart and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 means for performing the function specified by the flowchart and / or block diagram block or blocks.

[0134] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 means for performing the function specified by the flowchart and / or block diagram block or blocks.

[0135] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 means for performing the function specified by the flowchart and / or block diagram block or blocks.

[0136] Although preferred embodiments of the application have been described, additional modifications and changes can occur to others skilled in the art upon reading the preceding description.

[0137] It will be apparent to those skilled in the art that various modifications and variations can be made to the present application without departing from the spirit or scope of the application. Thus, it is intended that the present application cover modifications and variations of this application provided they come within the scope of the application and their equivalent technology.

Claims

1. A method for intelligent push based on dynamic interest decay analysis, characterized in that, The method comprises: When a target user triggers a video recommendation request, determining a first recommended video list based on the video recommendation request and the historical viewing records of the target user; Obtaining the first list viewing features of the target user on the first recommended video list, performing dynamic interest decay analysis based on the first list viewing features, and determining a second recommended video list according to the interest decay analysis result; Recommending the second recommended video list to the user end of the target user; Obtaining the first list viewing features of the target user on the first recommended video list, performing dynamic interest decay analysis based on the first list viewing features, and determining a second recommended video list according to the interest decay analysis result, comprising: Continuously recording the viewing time length of the target user watching each video in the first recommended video list to obtain the first list viewing features; Performing user interest analysis on the first list viewing features to obtain the current interest features of the user; Calling the historical interest features of the user, combining the current interest features of the user to perform dynamic interest decay analysis, and determining a second recommended video list according to the interest decay analysis result; Performing user interest analysis on the first list viewing features to obtain the current interest features of the user, comprising: Obtaining the first user interest category from multiple user interest categories; Calculating the viewing time length average of all videos in the first recommended video list belonging to the first user interest category to obtain the first category viewing time length average; Obtaining the viewing time length average of the remaining user interest categories in the same way as obtaining the first category viewing time length average to obtain the viewing time length average of multiple user interest categories as the current interest features of the user.

2. The method of claim 1, wherein, The target user triggers the video recommendation request when clicking any video in the short video application to enter the video playing interface.

3. The method of claim 1, wherein, Based on the video recommendation request and the historical viewing records of the target user, determining a first recommended video list, comprising: Connecting the full video pool according to the video recommendation request, wherein the full video pool contains all the recommendable videos; Extracting historical viewing features according to the historical viewing records of the target user, and determining multiple user interest categories and interest weights of each user interest category based on the historical viewing features; Based on multiple user interest categories and interest weights of each user interest category, video selection is performed in the full video pool to obtain the first recommended video list.

4. The method of claim 3, wherein, According to the historical viewing records of the target user, extracting historical viewing features, and determining multiple user interest categories and interest weights of each user interest category based on the historical viewing features, comprising: Traversing the historical viewing records to extract the first historical viewing record; When the viewing time length of the first historical viewing record exceeds the preset time threshold, obtaining the first video category of the first historical viewing record; According to the first video category, the occurrence frequency of the first video category in the full video category is incremented by 1; After processing the historical viewing records, the video categories with a non-zero occurrence frequency in the full video category are taken as the multiple user interest categories of the target user; The ratio of the occurrence frequency of each user interest category to the sum of occurrence frequencies of all user interest categories is calculated as the interest weight of the corresponding user interest category.

5. The method of claim 4, wherein, Based on the plurality of user interest categories and the interest weights of the user interest categories, video selection is performed in the full video pool to obtain the first recommended video list, including: Obtaining a preset recommended list video number; According to the preset recommended list video number and the interest weights of the user interest categories, obtaining a preset recommended number of each user interest category; Based on the preset recommended number of each user interest category, video is randomly selected in the corresponding video category in the full video pool to obtain the first recommended video list.

6. The method of claim 1, wherein, The user historical interest features are retrieved, and dynamic interest decay analysis is performed in combination with the current interest features of the user, and a second recommended video list is determined according to the interest decay analysis result, including: The user historical interest features are retrieved, and the user historical interest features include the historical viewing time average of each user interest category; The current user interest category viewing time average is obtained by traversing the current interest features of the user, and the historical viewing time average of the current user interest category is obtained in the user historical interest features; The ratio of the current user interest category viewing time average to the historical viewing time average is calculated to obtain the interest adjustment coefficient of the current user interest category; According to the manner of obtaining the interest adjustment coefficient of the current user interest category, the interest adjustment coefficient of each user interest category is obtained; Based on the interest adjustment coefficient of each user interest category, the corresponding interest weight is adjusted to obtain the adjusted interest weight of each user interest category as the interest decay analysis result; The second recommended video list is determined according to the adjusted interest weight of each user interest category.

7. An intelligent push system based on dynamic interest decay analysis characterized in that, For performing the method of any one of claims 1-6, including: A first recommendation module for determining a first recommended video list based on a video recommendation request and historical viewing records of a target user when the target user triggers the video recommendation request; An interest decay analysis module for obtaining first list viewing features of the target user on the first recommended video list, performing dynamic interest decay analysis based on the first list viewing features, and determining a second recommended video list according to the interest decay analysis result; A second recommendation module for recommending the second recommended video list to a user end of the target user.

Images