CN-122019884-A - User interest grading method and device, storage medium and electronic equipment

CN122019884ACN 122019884 ACN122019884 ACN 122019884ACN-122019884-A

Abstract

The application relates to a grading method and device for user interests, a storage medium and electronic equipment. The method comprises the steps of obtaining historical behavior data of a user, wherein the historical behavior data comprise the latest behavior time, the video watching completion degree, the video watching breadth, the behavior breadth and the negative feedback behavior of the user aiming at each interest, carrying out quantitative calculation on each interest of the user according to the historical behavior data and a preset interest weight calculation function to obtain a corresponding interest weight score, calculating an interest mean value and an interest standard deviation of interest distribution of the user based on all the interest weight scores of the user, and determining an interest grade of each interest of the user according to the interest mean value, the interest standard deviation and the corresponding interest weight score. The application solves the technical problem that the existing personalized recommendation distribution strategy is difficult to adapt to the interest distribution difference of different users, so that the interest classification is inaccurate.

Inventors

DU HONGGUANG
XU XIAOLE
ZHANG QUNFANG
HE KAI
ZHANG SHIJUN

Assignees

北京奇艺世纪科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260205

Claims (10)

1. A method of ranking user interests, comprising: Acquiring historical behavior data of a user, wherein the historical behavior data comprises the latest behavior time, the video watching completion degree, the video watching breadth, the behavior breadth and negative feedback behaviors of the user aiming at each interest; According to the historical behavior data and a preset interest weight calculation function, carrying out quantitative calculation on each interest of the user to obtain a corresponding interest weight score; calculating an interest mean value and an interest standard deviation of the interest distribution of the user based on all the interest weight scores of the user; and determining the interest grade of each interest of the user according to the interest mean value, the interest standard deviation and the corresponding interest weight score, wherein the interest grade comprises a strong interest, a medium interest and a weak interest.
2. The method of claim 1, wherein prior to performing a quantization calculation on each interest of the user to obtain a corresponding interest weight score, the method further comprises: acquiring group statistical data of each content channel of the user, wherein the group statistical data comprises the video watching completion degree of each interest of the user aiming at the corresponding content channel; for each content channel, determining a corresponding effective film watching completion degree threshold according to corresponding group statistical data; And under the condition that the watching completion degree of the user aiming at the target interests is lower than the effective watching completion degree threshold value of the content channels to which the target interests belong, identifying the target interests as invalid interests and filtering, wherein the target interests are any interests in the historical behavior data.
3. The method of claim 1, wherein quantitatively calculating each interest of the user based on the historical behavioral data and a preset interest weight calculation function, the obtaining a corresponding interest weight score comprises: determining the unprocessed interests as current interests, and performing the following processing on the current interests: Setting corresponding weights for the user aiming at the current interest film watching completion degree, the current interest film watching breadth and the current interest behavior breadth, and carrying out weighted summation to obtain an intermediate value; performing time attenuation processing on the latest behavior time of the user aiming at the current interest to obtain the time attenuation of the current interest; Calculating the product of the intermediate value and the time attenuation to obtain the interest weight score of the current interest; and identifying the current interest as invalid interest and filtering under the condition that the user has negative feedback action aiming at the current interest.
4. The method of claim 1, wherein calculating an interest mean and an interest standard deviation of the user's interest distribution based on all interest weight scores of the user comprises: Constructing an interest weight score set of the user according to the interest weight scores corresponding to all the interests of the user; And carrying out statistical analysis on the interest weight score set according to a preset statistical calculation rule, and calculating to obtain an interest mean value representing the overall interest level of the user and an interest standard deviation representing the user interest distribution discrete feature.
5. The method of claim 1, wherein for each interest of the user, determining its interest level from the interest mean, the interest standard deviation, and the corresponding interest weight score comprises: Acquiring an interest maximum value and an interest minimum value of interest distribution of the user based on all the interest weight scores of the user; Determining a group strong interest threshold and a group weak interest threshold according to the interest mean value, the interest standard deviation, the interest maximum value and the interest minimum value; determining the unprocessed interests as current interests, and performing the following processing on the current interests: Determining that the current interest is a strong interest when the interest weight score of the current interest is greater than or equal to the group strong interest threshold; Determining the current interest as a middle interest when the interest weight score of the current interest is smaller than the group strong interest threshold and larger than the group weak interest threshold; And determining that the current interest is a weak interest under the condition that the interest weight score of the current interest is smaller than or equal to the group weak interest threshold.
6. The method of claim 5, wherein determining a population strong interest threshold and a population weak interest threshold based on the mean of interest, the standard deviation of interest, the maximum of interest, and the minimum of interest comprises: Calculating the group strong interest threshold according to the interest mean value and the interest standard deviation: Wherein the said For the population strong interest threshold, the For the mean value of interest, the Is the standard deviation of interest; calculating the group weak interest threshold according to the interest mean value, the interest standard deviation, the interest maximum value and the interest minimum value: Wherein the said For the population weak interest threshold, the For the interest minimum value, the Is the maximum value of interest.
7. The method of any one of claims 1 to 6, wherein for each interest of the user, determining its interest level from the interest mean, the interest standard deviation, and the corresponding interest weight score comprises: Determining all interests of the user as strong interests in the case that the standard deviation of interests is equal to 0; And under the condition that the interest standard deviation is smaller than a preset standard deviation threshold value and the interest mean value is larger than an interest mean value threshold value, determining all interests of the user as strong interests, wherein the interest mean value threshold value is an interest weight score corresponding to a quarter span from the minimum value of the interest weight scores in the interest distribution of the user.
8. A user interest ranking apparatus, comprising: The system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring historical behavior data of a user, and the historical behavior data comprises the latest behavior time, the video watching completion degree, the video watching breadth, the behavior breadth and the negative feedback behavior of the user aiming at each interest; The first calculation module is used for carrying out quantitative calculation on each interest of the user according to the historical behavior data and a preset interest weight calculation function to obtain a corresponding interest weight score; The second calculation module is used for calculating an interest mean value and an interest standard deviation of the interest distribution of the user based on all the interest weight scores of the user; And the grading module is used for determining the interest grade of each interest of the user according to the interest mean value, the interest standard deviation and the corresponding interest weight score, wherein the interest grade comprises a strong interest, a medium interest and a weak interest.
9. A computer-readable storage medium, having stored thereon a computer program, characterized in that the computer program, when executed by a processor, performs the method of any of claims 1 to 7.
10. An electronic device comprising a memory and a processor, characterized in that the memory has stored therein a computer program, the processor being arranged to execute the method according to any of the claims 1 to 7 by means of the computer program.

Description

User interest grading method and device, storage medium and electronic equipment Technical Field The present application relates to the field of multimedia intelligent processing technologies, and in particular, to a method and apparatus for grading user interests, a storage medium, and an electronic device. Background In the field of long video personalized recommendation, the user interest intensity is an important basis for realizing refined content distribution and recommendation strategy regulation. In the prior art, the interest weight scores are calculated based on the behavior data of different contents or labels by quantitatively modeling the interests of the user, and the interests of the user are classified into different grades such as strong interests, medium interests, weak interests and the like according to the interest weight scores, so that a differential recommendation strategy is implemented in links such as content recall, sequencing or exposure control. The existing interest grading method adopts two types, namely one type is to select a fixed number of interests as strong interests after sorting according to the interest weight scores, and the other type is to divide the interest weight scores in a segmented mode by adopting a preset fixed threshold value. However, the user interest weight can be dynamically adjusted with the continuous change of the user behavior and the time decay mechanism, and different users have significant differences in the interest amount, the behavior frequency, the watching rhythm, the using time period and the like. For example, the interest coverage of partial users is wider and the activity is higher, and the interest quantity of partial users is smaller or the behavior is sparse, and for example, the interest weight of users with shorter watching time or slower watching rhythm is lower as a whole. The difference causes that the real interest distribution characteristics of different user individuals are difficult to accurately reflect by adopting a one-cut grading mode with fixed quantity or fixed threshold value, and the problem that strong interest identification is inaccurate or weak interest misjudgment easily occurs, so that personalized recommendation effect and user experience are affected. Disclosure of Invention The application provides a user interest grading method, a device, a storage medium and electronic equipment, which are used for solving the technical problem that the existing personalized recommendation distribution strategy is difficult to adapt to different user interest distribution differences, so that interest grading is inaccurate. The application provides a user interest grading method, which comprises the steps of obtaining historical behavior data of a user, wherein the historical behavior data comprise the latest behavior time, the movie watching completion degree, the movie watching breadth, the behavior breadth and the negative feedback behavior of the user aiming at each interest, carrying out quantitative calculation on each interest of the user according to the historical behavior data and a preset interest weight calculation function to obtain a corresponding interest weight score, calculating an interest mean value and an interest standard deviation of interest distribution of the user based on all the interest weight scores of the user, and determining an interest grade of the user according to the interest mean value, the interest standard deviation and the corresponding interest weight score aiming at each interest of the user, wherein the interest grade comprises a strong interest, a medium interest and a weak interest. The application provides a user interest grading device which comprises a first acquisition module, a first calculation module and a second calculation module, wherein the first acquisition module is used for acquiring historical behavior data of a user, the historical behavior data comprise the latest behavior time, the movie watching completion degree, the movie watching breadth and the negative feedback behavior of each interest of the user, the first calculation module is used for carrying out quantitative calculation on each interest of the user according to the historical behavior data and a preset interest weight calculation function to obtain a corresponding interest weight score, the second calculation module is used for calculating an interest mean value and an interest standard deviation of an interest distribution of the user according to all the interest weight scores of the user, and the grading module is used for determining an interest grade of each interest of the user according to the interest mean value, the interest standard deviation and the corresponding interest weight score, and the interest grade comprises a strong interest, a medium interest and a weak interest. As an optional example, the device further comprises a second acquisition module, a determination module and a f