CN114036403B - User interest detection method, device and storage medium - Google Patents
User interest detection method, device and storage medium Download PDFInfo
- Publication number
- CN114036403B CN114036403B CN202210012104.9A CN202210012104A CN114036403B CN 114036403 B CN114036403 B CN 114036403B CN 202210012104 A CN202210012104 A CN 202210012104A CN 114036403 B CN114036403 B CN 114036403B
- Authority
- CN
- China
- Prior art keywords
- interest
- interest point
- point
- probability value
- detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Abstract
The invention provides a user interest detection method, a user interest detection device and a storage medium, wherein the method comprises the following steps: obtaining interest distribution of a current user; the interest distribution comprises probability values of all interest points of the current user; sampling a preset number of interest points based on interest distribution and detection conditions; acquiring detection contents corresponding to a preset number of interest points, and pushing the detection contents to a recommendation page of a current user to realize a round of interest detection; when the current user refreshes the recommendation page, the probability value of a first interest point corresponding to the detection content in the interest distribution is updated based on the feedback behavior of the current user on the detection content displayed in the recommendation page before refreshing, and the probability values of other second interest points related to the first interest point corresponding to the detection content in the interest distribution are updated based on the relevance between the interest points. The invention improves the comprehensiveness and accuracy of interest detection of the cold start user and the efficiency of interest detection.
Description
Technical Field
The present invention relates to the field of information processing technologies, and in particular, to a method, an apparatus, and a storage medium for detecting user interest.
Background
In the personalized recommendation scenario, the new users are a relatively special class of users: firstly, the new user lacks clear understanding of the community and does not know what contents are in the community; secondly, the interest portraits of the new users are very sparse, making it difficult for the recommender system to infer what the users like. Meanwhile, a new user group is the main driving force for the continuous growth of the community, and the product experience and the subsequent activity retention degree of the new user group are very important for the community. Therefore, doing a cold start for a new user becomes an important and challenging task.
From the perspective of recommendation technology, one core is the interest detection for new users. By recommending rich contents in various fields to a new user, the new user can know the rich contents in a community, and meanwhile, feedback of the user is collected, so that the interest portrayal of the user is gradually enriched, and powerful assistance is provided for subsequent personalized recommendation. For interest detection of cold-start users, many methods are proposed in the industry, such as thompson sampling, UCB, etc. The methods sample all possible interest points of a user by using a certain sampling mode and select a part of the interest points to detect. However, these approaches sample the points of interest to be detected from all possible points of interest, and the interest detection process is not efficient and accurate enough.
Disclosure of Invention
The invention provides a user interest detection method, a user interest detection device and a storage medium, which are used for solving the defects of low efficiency and insufficient accuracy in a detection process in the prior art.
The invention provides a user interest detection method, which comprises the following steps:
obtaining interest distribution of a current user; wherein, the interest distribution comprises the probability value of the interest of the current user to all interest points;
sampling a preset number of interest points based on the interest distribution and the detection condition;
acquiring detection contents corresponding to the preset number of interest points, and pushing the detection contents to a recommendation page of the current user to realize a round of interest detection;
when the current user refreshes the recommended page, updating the probability value of a first interest point corresponding to the feedback behavior in the interest distribution based on the feedback behavior of the current user on the detection content displayed in the recommended page before refreshing, and updating the probability value of a second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the relevance between the interest points.
According to the user interest detection method provided by the invention, the relevance between the interest points comprises the conditional probability that the content of the interest point associated with any interest point is fed back positively after the content of the interest point is fed back positively, and the conditional probability that the content of the interest point associated with any interest point is fed back positively after the content of the interest point is fed back negatively; the conditional probability is calculated based on the feedback behavior of the sampling user to the content corresponding to each interest point.
According to the user interest detection method provided by the present invention, when the feedback behavior is a forward feedback, the updating of the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and the updating of the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the correlation between the interest points specifically include:
updating the probability value of the first interest point based on the information quantity fed back by the first interest point in the forward direction, the exposure times of the first interest point and the difference between the conditional probability that the content corresponding to the first interest point is fed back in the forward direction again and the probability value of the first interest point when the feedback behavior is the forward feedback;
updating the probability value of the second interest point based on the information quantity fed back by the first interest point in the forward direction, the exposure times of the second interest point and the difference between the conditional probability that the content corresponding to the second interest point is fed back in the forward direction and the probability value of the second interest point when the feedback behavior is the forward feedback.
According to the user interest detection method provided by the present invention, when the feedback behavior is a forward feedback, the updating of the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and the updating of the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the correlation between the interest points specifically include:
s(X)=s(X)+beta1(X)*f(ShowNum_X)*idf(A_pos)*(p(X|A)-s(X))
wherein, a is a first interest point, X is a first interest point or a second interest point, s (X) is a probability value of the interest point X, beta1(X) is a preset fixed parameter, shownnum _ X is the exposure times of the interest point X, f (shownnum _ X) is a monotonically decreasing function of the shownnum _ X, idf (a _ pos) is an information amount that the content of the interest point a is fed back in the forward direction, and p (X | a) is a conditional probability that the content corresponding to the interest point X is fed back in the forward direction when the feedback behavior is the forward feedback.
According to the user interest detection method provided by the present invention, when the feedback behavior is negative feedback, the updating of the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and the updating of the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the relevance between the interest points specifically include:
updating the probability value of the first interest point based on the information quantity fed back by the first interest point in a negative direction, the exposure times of the first interest point and the difference between the conditional probability of the content corresponding to the first interest point fed back by the first interest point in the negative direction and the probability value of the first interest point when the feedback behavior is negative feedback;
updating the probability value of the second interest point based on the information amount fed back by the first interest point in the negative direction, the exposure times of the second interest point and the difference between the conditional probability that the content corresponding to the second interest point is fed back in the positive direction and the probability value of the second interest point when the feedback behavior is negative feedback.
According to the user interest detection method provided by the present invention, when the feedback behavior is negative feedback, the updating of the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and the updating of the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the relevance between the interest points specifically include:
s(X)=s(X)+beta2(X)*f(ShowNum_X)*idf(A_neg)*(p(X|!A)-s(X))
wherein, a is a first interest point, X is the first interest point or a second interest point, s (X) is a probability value of the interest point X, beta2(X) is a preset fixed parameter, shownnum _ X is the exposure frequency of the interest point X, f (shownnum _ X) is a monotonically decreasing function of the shownnum _ X, idf (a _ neg) is an information amount of the interest point a whose content is fed back in the negative direction, and p (X | a) is a conditional probability that the content corresponding to the interest point X is fed back in the positive direction when the feedback behavior is negative feedback.
According to the user interest detection method provided by the invention, the sampling of a preset number of interest points based on the interest distribution and detection condition specifically comprises the following steps:
determining the sampling probability of each interest point based on the probability value of each interest point in the interest distribution and the current exposure times of each interest point; the sampling probability of any interest point is positively correlated with the probability value of any interest point in interest distribution, and is negatively correlated with the current exposure times of any interest point;
a preset number of interest points are sampled based on the sampling probability of each interest point.
According to the user interest detection method provided by the invention, the sampling probability of each interest point is the product of the probability value of each interest point in interest distribution and a detection coefficient;
wherein the detection coefficient is smaller as the number of times of detection of any interest point is larger.
The present invention also provides a user interest detection apparatus, comprising:
the interest distribution acquisition unit is used for acquiring the interest distribution of the current user; wherein, the interest distribution comprises the probability value of the interest of the current user to all interest points;
the interest point sampling unit is used for sampling a preset number of interest points based on the interest distribution and detection condition;
the detection content pushing unit is used for acquiring detection contents corresponding to the preset number of interest points, pushing the detection contents to a recommendation page of the current user and realizing a round of interest detection;
when the current user refreshes the recommended page, updating the probability value of a first interest point corresponding to the feedback behavior in the interest distribution based on the feedback behavior of the current user on the detection content displayed in the recommended page before refreshing, and updating the probability value of a second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the relevance between the interest points.
The present invention also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the steps of the user interest detection method according to any of the above.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when being executed by a processor, carries out the steps of the user interest detection method according to any one of the above-mentioned claims.
According to the user interest detection method, the device and the storage medium, the probability value of the first interest point corresponding to the feedback behavior in the interest distribution of the user is updated through the feedback behavior of the user on the detection content displayed in the recommended page, the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution is updated based on the relevance between the interest points, the information amount of each behavior of the user is fully utilized, the comprehensiveness and accuracy of updating of the interest distribution are improved, a perfect interest portrait can be established for the user as soon as possible, then the interest points are sampled based on the probability values of the interest points in the interest distribution, interest detection is achieved, and the interest detection efficiency of the cold-start user is improved.
Drawings
In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed for the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
FIG. 1 is a schematic flow chart of a user interest detection method provided by the present invention;
FIG. 2 is a schematic diagram of the correlation between points of interest provided by the present invention;
FIG. 3 is a schematic diagram of a click propagation matrix provided by the present invention;
FIG. 4 is a schematic diagram of a non-click conductance matrix provided by the present invention;
FIG. 5 is a schematic structural diagram of a user interest detection apparatus provided in the present invention;
fig. 6 is a schematic structural diagram of an electronic device provided in the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a schematic flow chart of a user interest detection method according to an embodiment of the present invention, as shown in fig. 1, the method includes:
when the current user refreshes the recommended page, updating the probability value of a first interest point corresponding to the feedback behavior in the interest distribution based on the feedback behavior of the current user on the detection content displayed in the recommended page before refreshing, and updating the probability value of a second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the relevance between the interest points.
Specifically, the interest distribution of the current user includes probability values that the user is interested in all preset interest points at present. The preset interest points can be divided into themes according to the content in the community, so that the interest points with proper granularity are set, and the preset interest points can comprehensively cover all fields, such as emotion, fashion, mother and infant and the like. When the current user is a new user, an initial value may be set for his interest distribution. Here, a relatively reasonable initial interest distribution may be given to the cold-start user based on a priori knowledge to enhance subsequent detection efficiency. The number of people under each interest point can be analyzed and counted according to the community big data, and the initial probability of each interest point is calculated. Here, the initial probability of the popular interest point may be set higher, and the initial probability of the less popular interest point may be set lower.
Then, according to the probability value corresponding to each interest point in the current interest distribution of the user and the current detection condition, such as the detected turn, the detection content pushed to the user by each detection, etc., the interest points which have higher probability value of preset quantity and are possibly more interested by the current user are sampled from the probability value for interest detection. And acquiring detection contents corresponding to the preset number of interest points, and pushing the detection contents to a recommendation page of the current user to realize one round of interest detection. The corresponding high-quality content can be selected for each interest point in an off-line manner, an index is established, and the detection content corresponding to the interest point can be directly obtained according to the index when the detection content is obtained.
When the current user browses the current content in the recommended page and refreshes the recommended page, the interest distribution of the user can be updated according to the feedback behavior of the user on the detected content displayed in the recommended page in the last refreshing, so that the interest images of the user are enriched, and the interest distribution of the user is constructed more accurately. Specifically, the probability value of the first interest point corresponding to the feedback behavior may be updated based on the feedback behavior of the current user on the detected content in the last brushing. Wherein the feedback behavior may include positive feedback, such as clicking, previewing, viewing, etc., indicating that the user has shown an interest in the content/interest point; feedback behavior may also include negative feedback, such as a certain content exposed for the recommended page, a user not clicking or folding or clicking uninteresting, etc., indicating that the user is not interested in the content/point of interest representation. The interest point corresponding to the content generating the feedback behavior is the first interest point, and the interest point which is associated with the first interest point is the second interest point.
Here, for a first interest point directly generating a feedback behavior, the probability value of the feedback behavior in the interest distribution may be updated according to whether the feedback behavior is positive feedback or negative feedback. For example, if the feedback is positive, the probability value of the corresponding first interest point is adjusted to be high, and if the feedback is negative, the probability value of the corresponding first interest point is adjusted to be low.
It should be noted that, an ideal interest detection method should have both breadth-first and depth-first, the breadth detection is comprehensive, and the depth detection is accurate, so as to implement an efficient detection process. The comprehensiveness and accuracy of interest distribution updating are important prerequisites for ensuring the comprehensiveness and accuracy of interest detection. Therefore, in order to improve the comprehensiveness and accuracy of updating the interest distribution, a perfect interest portrait is established for the user as soon as possible, so that the interest detection efficiency is improved, and the probability value of the second interest point associated with the first interest point can be updated by fully utilizing the information amount of the feedback action of the current user in the previous brushing.
Wherein, the relevance between the interest points can be established in advance, and the relevance atmosphere has positive relevance and negative relevance. The positive relevance indicates that the two topics are close to each other, and the number of the common occurrences in the interest point data of the sampling user is large; negative correlation indicates that the two subjects are opposite, and the number of co-occurrences in the interest point data of the sampling user is very small. Here, according to a large batch of sampling interest points and interest points which are not interested by the users, co-occurrence data of any two interest points on the users can be counted, so that the relevance between any two interest points can be determined. If the positive association between any first interest point and any second interest point is strong and the number of co-occurrences is large, the adjustment direction of the second interest point may be determined according to the adjustment direction of the probability value of the first interest point, for example, the probability values are all increased or all decreased. If the negative association between any first interest point and any second interest point is strong and the number of co-occurrences is very small, the adjustment direction of the second interest point can be reversely determined according to the adjustment direction of the probability value of the first interest point, for example, the probability value of one interest point is increased, and the probability value of another interest point is decreased. As shown in fig. 2, after the probe contents of several interest points are recommended for the current user, the user clicks the military contents, so that the interest level of the user in the first interest point "military" and the second interest point "political" which is positively correlated is increased, and the corresponding probability value is increased, and meanwhile, the interest level in the second interest point "fashion" and "mother and infant" which are negatively correlated with "military" is decreased, and the corresponding probability value is decreased.
When the user refreshes the recommendation page and updates the interest distribution according to the mode, the steps can be repeated to start a new round of interest detection until a perfect interest portrait of the user is established.
According to the method provided by the embodiment of the invention, the probability value of the first interest point corresponding to the feedback behavior in the interest distribution of the user is updated through the feedback behavior of the user on the last brushing of the detection content displayed in the recommended page, the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution is updated based on the relevance between the interest points, the information amount of each behavior of the user is fully utilized, the comprehensiveness and the accuracy of updating of the interest distribution are improved, so that a perfect interest portrait can be established for the user as soon as possible, then the interest points are sampled based on the probability values of the interest points in the interest distribution, the interest detection is realized, and the interest detection efficiency of the cold start user is improved.
Based on the above embodiment, the relevance between the interest points includes a conditional probability that, after the content of any interest point is fed back in a positive direction, the content of the interest point associated with the interest point is also fed back in the positive direction, and a conditional probability that, after the content of any interest point is fed back in a negative direction, the content of the interest point associated with the interest point is fed back in the positive direction; the conditional probability is calculated based on the feedback behavior of the sampling user to the content corresponding to each interest point.
Specifically, the relevance between the interest points includes a conditional probability that the content of any interest point is fed back positively after the content of any interest point is fed back positively, that is, a positive relevance degree, and a conditional probability that the content of any interest point is fed back positively after the content of any interest point is fed back negatively, that is, a negative relevance degree. The conditional probability can be calculated based on the feedback behavior of the sampling user to the content corresponding to each interest point.
Here, the interest points of interest of each sampling user may be inferred based on the feedback behavior of each sampling user to the content corresponding to each interest point, so as to obtain the interest point list of each sampling user. For example, whether a sampling user is interested in any point of interest can be determined according to the click rate of the sampling user on the content of the point of interest. If the click rate is high, the sampling user can be considered to be interested in the interest point. For all the preset interest points, if a certain interest point is not in the interest point list of a certain sampling user, the sampling user can be considered to be not interested in the interest point. Then, the conditional probability can be calculated according to whether all the sampling users are interested in each interest point or not.
For example, assuming that there are an interest point a and an interest point X, a conditional probability P (X | a) that the content of the interest point X is fed back in the forward direction after the content of the interest point a is fed back in the forward direction can be calculated:
P(X|A) = P(A, X) / P(A)
wherein, P (A, X) is the ratio of the number of sampling users interested in both A and X to the total number of sampling users, and P (A) is the ratio of the number of sampling users interested in A to the total number of sampling users.
Similarly, a conditional probability P (X | a) that the content of the point of interest X is fed back positively after the content of the point of interest a is fed back negatively can also be calculated:
P(X|!A) = P(!A, X) / P(!A)
wherein, P (! A, X) is the ratio of the number of sampling users which are not interested in A but are interested in X to the total number of sampling users, and P (! A) is the ratio of the number of sampling users which are not interested in A to the total number of sampling users.
After the conditional probability that the content of any interest point is fed back in the forward direction and the content of any other interest point is also fed back in the forward direction is calculated, all the conditional probabilities can be organized into a click conduction matrix; after the conditional probability that the content of any interest point is fed back in the negative direction and the content of any other interest point is fed back in the positive direction is calculated, all the conditional probabilities can be organized into a non-click conducting matrix. Wherein, for the conditional probability smaller than the preset threshold, it may be set to 0. The click conductance matrix and the no-click conductance matrix are shown in fig. 3 and 4.
Based on any of the above embodiments, when the feedback behavior is forward feedback, the updating the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and based on the relevance between the interest points, the updating the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution specifically includes:
updating the probability value of the first interest point based on the information quantity fed back by the first interest point in the forward direction, the exposure times of the first interest point and the difference between the conditional probability that the content corresponding to the first interest point is fed back in the forward direction again and the probability value of the first interest point when the feedback behavior is the forward feedback;
updating the probability value of the second interest point based on the information quantity fed back by the first interest point in the forward direction, the exposure times of the second interest point and the difference between the conditional probability that the content corresponding to the second interest point is fed back in the forward direction and the probability value of the second interest point when the feedback behavior is the forward feedback.
Specifically, when a feedback behavior of a current user for a certain probe content is a forward feedback, a first interest point corresponding to the feedback behavior and a second interest point associated with the first interest point are determined.
Then, the probability value of the first interest point may be updated based on the amount of information that the first interest point is fed back in the forward direction, the number of times of exposure of the first interest point, and a difference between the conditional probability that the content corresponding to the first interest point is fed back in the forward direction again and the probability value of the first interest point when the feedback behavior is the forward feedback. The information quantity of the first interest point fed back in the forward direction is inversely related to the popularity of the first interest point, the higher the popularity is, the lower the information quantity of the first interest point fed back in the forward direction is, and the smaller the adjustment amplitude of the probability value of the first interest point is. The popularity of any point of interest may be determined based on the ratio of the number of sampled users interested in the point of interest to the total number of sampled users. In addition, the number of times of exposure of the first interest point is the number of times of historical pushing of the content corresponding to the first interest point. The higher the exposure times of the first interest point, the more confidence the current probability value is, and the smaller the adjustment amplitude of the probability value is.
The probability value of the second interest point can be updated based on the information amount fed back forward by the first interest point, the exposure times of any second interest point, and the difference between the conditional probability that the content corresponding to the second interest point is fed back forward and the probability value of the second interest point when the feedback behavior is the forward feedback. Similarly, the lower the information amount fed back forward by the first interest point is, the smaller the adjustment amplitude of the probability value of the second interest point is. The higher the exposure times of the second interest point, the smaller the adjustment amplitude of the probability value of this time. In addition, when the content of the first interest point is fed back in the forward direction, the higher the conditional probability that the content corresponding to the second interest point is fed back in the forward direction is, which indicates that the degree of positive association between the first interest point and the second interest point is higher, the higher the reference value of the feedback behavior for the first interest point to the second interest point is, and therefore the larger the adjustment amplitude of the probability value is.
Based on any of the above embodiments, when the feedback behavior is forward feedback, the updating the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and based on the relevance between the interest points, the updating the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution specifically includes:
s(X)=s(X)+beta1(X)*f(ShowNum_X)*idf(A_pos)*(p(X|A)-s(X))
wherein, a is a first interest point, X is a first interest point or a second interest point, s (X) is a probability value of the interest point X, beta1(X) is a preset fixed parameter, shownnum _ X is the exposure times of the interest point X, f (shownnum _ X) is a monotonically decreasing function of the shownnum _ X, idf (a _ pos) is an information amount that the content of the interest point a is fed back in the forward direction, and p (X | a) is a conditional probability that the content corresponding to the interest point X is fed back in the forward direction when the feedback behavior is the forward feedback.
Here, beta (X) is a fixed parameter, which can be understood as an updated base step size. f (ShowNum _ X) is a monotonically decreasing function of ShowNum _ X (the number of exposures of the interest point X), and the step size of probability value updating is adjusted through the function, wherein the larger the ShowNum _ X is, the smaller f (ShowNum _ X) is, and the smaller the step size of probability value updating is. idf (a _ pos) is the amount of information that the content of point of interest a is fed back forward, which may be the difference between the value 1 and the popularity of point of interest a.
Based on any of the above embodiments, when the feedback behavior is negative feedback, the updating the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and based on the relevance between the interest points, the updating the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution specifically includes:
updating the probability value of the first interest point based on the information quantity fed back by the first interest point in a negative direction, the exposure times of the first interest point and the difference between the conditional probability of the content corresponding to the first interest point fed back by the first interest point in the negative direction and the probability value of the first interest point when the feedback behavior is negative feedback;
updating the probability value of the second interest point based on the information amount fed back by the first interest point in the negative direction, the exposure times of the second interest point and the difference between the conditional probability that the content corresponding to the second interest point is fed back in the positive direction and the probability value of the second interest point when the feedback behavior is negative feedback.
Specifically, when a feedback behavior of a current user for a certain detection content is negative feedback, a first interest point corresponding to the feedback behavior and a second interest point associated with the first interest point are determined.
Then, the probability value of the first interest point may be updated based on the information amount negatively fed back of the first interest point, the number of times of exposure of the first interest point, and a difference between a conditional probability that content corresponding to the first interest point is positively fed back next time and the probability value of the first interest point when the feedback behavior is negative feedback. The negatively fed back information quantity of the first interest point is positively correlated with the popularity of the first interest point, and the higher the popularity is, the higher the information quantity of the first interest point is negatively fed back is, and the larger the adjustment amplitude of the probability value of the first interest point is. In addition, the higher the exposure times of the first interest point, the more confidence the current probability value is, the smaller the adjustment amplitude of the probability value is.
The probability value of the second interest point may also be updated based on the information amount negatively fed back by the first interest point, the exposure times of any one of the second interest points, and the difference between the conditional probability that the content corresponding to the second interest point is positively fed back and the probability value of the second interest point when the feedback behavior is negative feedback. Similarly, the lower the information amount negatively fed back by the first interest point is, the smaller the adjustment amplitude of the probability value of the second interest point is. The higher the exposure times of the second interest point, the smaller the adjustment amplitude of the probability value of this time. In addition, when the content of the first interest point is fed back negatively, the higher the conditional probability that the content corresponding to the second interest point is fed back positively is, which indicates that the degree of negative association between the first interest point and the second interest point is higher, the higher the reference value of the feedback behavior for the first interest point to the second interest point is, and therefore the larger the adjustment amplitude of the probability value is.
Based on any of the above embodiments, when the feedback behavior is negative feedback, the updating the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and based on the relevance between the interest points, the updating the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution specifically includes:
s(X)=s(X)+beta2(X)*f(ShowNum_X)*idf(A_neg)*(p(X|!A)-s(X))
wherein, a is a first interest point, X is the first interest point or a second interest point, s (X) is a probability value of the interest point X, beta2(X) is a preset fixed parameter, shownnum _ X is the exposure frequency of the interest point X, f (shownnum _ X) is a monotonically decreasing function of the shownnum _ X, idf (a _ neg) is an information amount of the interest point a whose content is fed back in the negative direction, and p (X | a) is a conditional probability that the content corresponding to the interest point X is fed back in the positive direction when the feedback behavior is negative feedback.
Here, beta2(X) is a fixed parameter and may also be understood as an updated base step size. Considering that the amount of information fed back in the positive direction is larger than that fed back in the negative direction, beta1(X) > beta2(X) may be set. f (ShowNum _ X) is a monotonically decreasing function of ShowNum _ X (the number of exposures of the interest point X), and the step size of probability value updating is adjusted through the function, wherein the larger the ShowNum _ X is, the smaller f (ShowNum _ X) is, and the smaller the step size of probability value updating is. idf (a _ neg) is the amount of information that the content of the point of interest a is negatively fed back, which may be equal to the popularity of the point of interest a.
Based on any of the above embodiments, step 120 specifically includes:
determining the sampling probability of each interest point based on the probability value of each interest point in the interest distribution and the current exposure times of each interest point; the sampling probability of any interest point is positively correlated with the probability value of any interest point in interest distribution, and is negatively correlated with the current exposure times of any interest point;
a preset number of interest points are sampled based on the sampling probability of each interest point.
Specifically, when selecting an interest point for interest detection, the balance of the E & E problem, i.e. the balance between utilization (Explore) and exploration (Explore), needs to be grasped. Therefore, the sampling probability of each interest point can be comprehensively determined based on the probability value of each interest point in the interest distribution and the current exposure times of each interest point. The sampling probability of any interest point is positively correlated with the probability value of the interest point in the interest distribution, and is negatively correlated with the current exposure times of the interest point. Here, the probability value of the interest point in the interest distribution represents "utilization", and the current exposure number of the interest point represents "exploration". The sampling probability of the interest points is set to be positively correlated with the probability value of the interest points in the interest distribution, the interest portrait of the user learned before can be fully utilized, the accuracy of interest exploration is improved, meanwhile, the sampling probability of the interest points and the current exposure times of the interest points are set to be negatively correlated, namely the interest points with less exposure times are sampled with higher probability, the user can be explored more and related to less interest points and pushed to the user, the comprehensiveness of the interest exploration is improved, and the balance of 'utilization' and 'exploration' is achieved.
The interest points are then sampled for interest detection based on the sampling probabilities of the respective interest points.
Based on any of the above embodiments, the sampling probability of each interest point is a product of a probability value of each interest point in the interest distribution and a detection coefficient;
wherein the detection coefficient is smaller as the number of times of detection of any interest point is larger.
In particular, diversity control may also be introduced when sampling points of interest. Wherein, the principle of diversity control is as follows: in the early stage of interest detection, the diversity of interest points can be enhanced, the comprehensiveness of interest detection is enlarged, along with the detection depth, the user is more and more informed, the diversity control can be gradually relaxed, and the detected interest points are focused more accurately.
Here, the number of rounds of detection of each interest point may be obtained, and the more rounds, the deeper the interest detection of the interest point is. Based on the round of interest point detection, the detection coefficients used when sampling the interest points are determined. The detection times of any interest point are less, the detection coefficient corresponding to the interest point is larger, and the interest point is easier to be sampled as the interest to be detected; correspondingly, the more times of detection of any interest point, the smaller the detection coefficient corresponding to the interest point, and the less easily the interest point is selected again for detection.
The following describes the user interest detection apparatus provided by the present invention, and the user interest detection apparatus described below and the user interest detection method described above may be referred to in correspondence with each other.
Based on any of the above embodiments, fig. 5 is a schematic structural diagram of a user interest detection apparatus provided in an embodiment of the present invention, as shown in fig. 5, the apparatus includes: an interest distribution obtaining unit 510, an interest point sampling unit 520, and a probe content pushing unit 530.
The interest distribution obtaining unit 510 is configured to obtain an interest distribution of a current user; wherein, the interest distribution comprises the probability value of the interest of the current user to all interest points;
the interest point sampling unit 520 is configured to sample a preset number of interest points based on the interest distribution and the detection condition;
the detection content pushing unit 530 is configured to obtain detection contents corresponding to the preset number of interest points, and push the detection contents to a recommendation page of the current user to implement a round of interest detection;
when the current user refreshes the recommended page, updating the probability value of a first interest point corresponding to the feedback behavior in the interest distribution based on the feedback behavior of the current user on the detection content displayed in the recommended page before refreshing, and updating the probability value of a second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the relevance between the interest points.
According to the device provided by the embodiment of the invention, the probability value of the first interest point corresponding to the feedback behavior in the interest distribution of the user is updated through the feedback behavior of the user on the last brushing of the detection content displayed in the recommended page, the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution is updated based on the relevance between the interest points, the information amount of each behavior of the user is fully utilized, the comprehensiveness and the accuracy of updating of the interest distribution are improved, so that a perfect interest portrait can be established for the user as soon as possible, then the interest points are sampled based on the probability values of the interest points in the interest distribution, the interest detection is realized, and the interest detection efficiency of the cold start user is improved.
Based on any of the above embodiments, the relevance between the interest points includes a conditional probability that, after the content of any interest point is fed back in a positive direction, the content of the interest point associated with the interest point is also fed back in the positive direction, and a conditional probability that, after the content of any interest point is fed back in a negative direction, the content of the interest point associated with the interest point is fed back in the positive direction; the conditional probability is calculated based on the feedback behavior of the sampling user to the content corresponding to each interest point.
Based on any of the above embodiments, when the feedback behavior is forward feedback, the updating the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and based on the relevance between the interest points, the updating the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution specifically includes:
updating the probability value of the first interest point based on the information quantity fed back by the first interest point in the forward direction, the exposure times of the first interest point and the difference between the conditional probability that the content corresponding to the first interest point is fed back in the forward direction again and the probability value of the first interest point when the feedback behavior is the forward feedback;
updating the probability value of the second interest point based on the information quantity fed back by the first interest point in the forward direction, the exposure times of the second interest point and the difference between the conditional probability that the content corresponding to the second interest point is fed back in the forward direction and the probability value of the second interest point when the feedback behavior is the forward feedback.
Based on any of the above embodiments, when the feedback behavior is forward feedback, the updating the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and based on the relevance between the interest points, the updating the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution specifically includes:
s(X)=s(X)+beta1(X)*f(ShowNum_X)*idf(A_pos)*(p(X|A)-s(X))
wherein, a is a first interest point, X is a first interest point or a second interest point, s (X) is a probability value of the interest point X, beta1(X) is a preset fixed parameter, shownnum _ X is the exposure times of the interest point X, f (shownnum _ X) is a monotonically decreasing function of the shownnum _ X, idf (a _ pos) is an information amount that the content of the interest point a is fed back in the forward direction, and p (X | a) is a conditional probability that the content corresponding to the interest point X is fed back in the forward direction when the feedback behavior is the forward feedback.
Based on any of the above embodiments, when the feedback behavior is negative feedback, the updating the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and based on the relevance between the interest points, the updating the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution specifically includes:
updating the probability value of the first interest point based on the information quantity fed back by the first interest point in a negative direction, the exposure times of the first interest point and the difference between the conditional probability of the content corresponding to the first interest point fed back by the first interest point in the negative direction and the probability value of the first interest point when the feedback behavior is negative feedback;
updating the probability value of the second interest point based on the information amount fed back by the first interest point in the negative direction, the exposure times of the second interest point and the difference between the conditional probability that the content corresponding to the second interest point is fed back in the positive direction and the probability value of the second interest point when the feedback behavior is negative feedback.
Based on any of the above embodiments, when the feedback behavior is negative feedback, the updating the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and based on the relevance between the interest points, the updating the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution specifically includes:
s(X)=s(X)+beta2(X)*f(ShowNum_X)*idf(A_neg)*(p(X|!A)-s(X))
wherein, a is a first interest point, X is the first interest point or a second interest point, s (X) is a probability value of the interest point X, beta2(X) is a preset fixed parameter, shownnum _ X is the exposure frequency of the interest point X, f (shownnum _ X) is a monotonically decreasing function of the shownnum _ X, idf (a _ neg) is an information amount of the interest point a whose content is fed back in the negative direction, and p (X | a) is a conditional probability that the content corresponding to the interest point X is fed back in the positive direction when the feedback behavior is negative feedback.
Based on any of the above embodiments, the sampling a preset number of interest points based on the interest distribution and detection condition specifically includes:
determining the sampling probability of each interest point based on the probability value of each interest point in the interest distribution and the current exposure times of each interest point; the sampling probability of any interest point is positively correlated with the probability value of any interest point in interest distribution, and is negatively correlated with the current exposure times of any interest point;
a preset number of interest points are sampled based on the sampling probability of each interest point.
Based on any of the above embodiments, the sampling probability of each interest point is a product of a probability value of each interest point in the interest distribution and a detection coefficient;
wherein the detection coefficient is smaller as the number of times of detection of any interest point is larger.
Fig. 6 illustrates a physical structure diagram of an electronic device, which may include, as shown in fig. 6: a processor (processor)610, a communication Interface (Communications Interface)620, a memory (memory)630 and a communication bus 640, wherein the processor 610, the communication Interface 620 and the memory 630 communicate with each other via the communication bus 640. The processor 610 may invoke logic instructions in the memory 630 to perform a user interest detection method comprising: obtaining interest distribution of a current user; wherein, the interest distribution comprises the probability value of the interest of the current user to all interest points; sampling a preset number of interest points based on the interest distribution and the detection condition; acquiring detection contents corresponding to the preset number of interest points, and pushing the detection contents to a recommendation page of the current user to realize a round of interest detection; when the current user refreshes the recommended page, updating the probability value of a first interest point corresponding to the feedback behavior in the interest distribution based on the feedback behavior of the current user on the detection content displayed in the recommended page before refreshing, and updating the probability value of a second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the relevance between the interest points.
In addition, the logic instructions in the memory 630 may be implemented in software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In another aspect, the present invention also provides a computer program product, the computer program product comprising a computer program, the computer program being stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, a computer is capable of executing the user interest detection method provided by the above methods, the method comprising: obtaining interest distribution of a current user; wherein, the interest distribution comprises the probability value of the interest of the current user to all interest points; sampling a preset number of interest points based on the interest distribution and the detection condition; acquiring detection contents corresponding to the preset number of interest points, and pushing the detection contents to a recommendation page of the current user to realize a round of interest detection; when the current user refreshes the recommended page, updating the probability value of a first interest point corresponding to the feedback behavior in the interest distribution based on the feedback behavior of the current user on the detection content displayed in the recommended page before refreshing, and updating the probability value of a second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the relevance between the interest points.
In yet another aspect, the present invention also provides a non-transitory computer-readable storage medium having stored thereon a computer program, which when executed by a processor, implements a method for detecting user interest provided by the above methods, the method comprising: obtaining interest distribution of a current user; wherein, the interest distribution comprises the probability value of the interest of the current user to all interest points; sampling a preset number of interest points based on the interest distribution and the detection condition; acquiring detection contents corresponding to the preset number of interest points, and pushing the detection contents to a recommendation page of the current user to realize a round of interest detection; when the current user refreshes the recommended page, updating the probability value of a first interest point corresponding to the feedback behavior in the interest distribution based on the feedback behavior of the current user on the detection content displayed in the recommended page before refreshing, and updating the probability value of a second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the relevance between the interest points.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (9)
1. A method for detecting user interest, comprising:
obtaining interest distribution of a current user; wherein, the interest distribution comprises the probability value of the interest of the current user to all interest points;
sampling a preset number of interest points based on the interest distribution and the detection condition;
acquiring detection contents corresponding to the preset number of interest points, and pushing the detection contents to a recommendation page of the current user to realize a round of interest detection;
when the current user refreshes the recommended page, updating the probability value of a first interest point corresponding to the feedback behavior in the interest distribution based on the feedback behavior of the current user on the detection content displayed in the recommended page before refreshing, and updating the probability value of a second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the relevance between the interest points;
the relevance between the interest points comprises the conditional probability that the content of the interest point associated with any interest point is fed back positively after the content of the interest point is fed back positively, and the conditional probability that the content of the interest point associated with any interest point is fed back positively after the content of the interest point is fed back negatively; the conditional probability is calculated based on the feedback behavior of the sampling user to the content corresponding to each interest point.
2. The method according to claim 1, wherein when the feedback behavior is a forward feedback, the updating the probability value of a first interest point corresponding to the feedback behavior in the interest distribution, and updating the probability value of a second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the association between the interest points specifically includes:
updating the probability value of the first interest point based on the information quantity fed back by the first interest point in the forward direction, the exposure times of the first interest point and the difference between the conditional probability that the content corresponding to the first interest point is fed back in the forward direction again and the probability value of the first interest point when the feedback behavior is the forward feedback;
updating the probability value of the second interest point based on the information quantity fed back by the first interest point in the forward direction, the exposure times of the second interest point and the difference between the conditional probability that the content corresponding to the second interest point is fed back in the forward direction and the probability value of the second interest point when the feedback behavior is the forward feedback.
3. The method according to claim 2, wherein when the feedback behavior is a forward feedback, the updating the probability value of a first interest point corresponding to the feedback behavior in the interest distribution, and updating the probability value of a second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the association between the interest points specifically includes:
s(X)=s(X)+beta1(X)*f(ShowNum_X)*idf(A_pos)*(p(X|A)-s(X))
wherein, a is a first interest point, X is a first interest point or a second interest point, s (X) is a probability value of the interest point X, beta1(X) is a preset fixed parameter, shownnum _ X is the exposure times of the interest point X, f (shownnum _ X) is a monotonically decreasing function of the shownnum _ X, idf (a _ pos) is an information amount that the content of the interest point a is fed back in the forward direction, and p (X | a) is a conditional probability that the content corresponding to the interest point X is fed back in the forward direction when the feedback behavior is the forward feedback.
4. The method according to claim 1, wherein when the feedback behavior is negative feedback, the updating the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and based on the association between the interest points, the updating the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution specifically includes:
updating the probability value of the first interest point based on the information quantity fed back by the first interest point in a negative direction, the exposure times of the first interest point and the difference between the conditional probability of the content corresponding to the first interest point fed back by the first interest point in the negative direction and the probability value of the first interest point when the feedback behavior is negative feedback;
updating the probability value of the second interest point based on the information amount fed back by the first interest point in the negative direction, the exposure times of the second interest point and the difference between the conditional probability that the content corresponding to the second interest point is fed back in the positive direction and the probability value of the second interest point when the feedback behavior is negative feedback.
5. The method according to claim 4, wherein when the feedback behavior is negative feedback, the updating the probability value of the first interest point corresponding to the feedback behavior in the interest distribution, and based on the association between the interest points, the updating the probability value of the second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution specifically includes:
s(X)=s(X)+beta2(X)*f(ShowNum_X)*idf(A_neg)*(p(X|!A)-s(X))
wherein, a is a first interest point, X is the first interest point or a second interest point, s (X) is a probability value of the interest point X, beta2(X) is a preset fixed parameter, shownnum _ X is the exposure frequency of the interest point X, f (shownnum _ X) is a monotonically decreasing function of the shownnum _ X, idf (a _ neg) is an information amount of the interest point a whose content is fed back in the negative direction, and p (X | a) is a conditional probability that the content corresponding to the interest point X is fed back in the positive direction when the feedback behavior is negative feedback.
6. The method according to any one of claims 1 to 5, wherein the sampling a preset number of interest points based on the interest distribution and detection conditions specifically includes:
determining the sampling probability of each interest point based on the probability value of each interest point in the interest distribution and the current exposure times of each interest point; the sampling probability of any interest point is positively correlated with the probability value of any interest point in interest distribution, and is negatively correlated with the current exposure times of any interest point;
a preset number of interest points are sampled based on the sampling probability of each interest point.
7. The method according to claim 6, wherein the sampling probability of each interest point is a product of a probability value of each interest point in an interest distribution and a detection coefficient;
wherein the detection coefficient is smaller as the number of times of detection of any interest point is larger.
8. A user interest detection apparatus, comprising:
the interest distribution acquisition unit is used for acquiring the interest distribution of the current user; wherein, the interest distribution comprises the probability value of the interest of the current user to all interest points;
the interest point sampling unit is used for sampling a preset number of interest points based on the interest distribution and detection condition;
the detection content pushing unit is used for acquiring detection contents corresponding to the preset number of interest points, pushing the detection contents to a recommendation page of the current user and realizing a round of interest detection;
when the current user refreshes the recommended page, updating the probability value of a first interest point corresponding to the feedback behavior in the interest distribution based on the feedback behavior of the current user on the detection content displayed in the recommended page before refreshing, and updating the probability value of a second interest point related to the first interest point corresponding to the feedback behavior in the interest distribution based on the relevance between the interest points;
the relevance between the interest points comprises the conditional probability that the content of the interest point associated with any interest point is fed back positively after the content of the interest point is fed back positively, and the conditional probability that the content of the interest point associated with any interest point is fed back positively after the content of the interest point is fed back negatively; the conditional probability is calculated based on the feedback behavior of the sampling user to the content corresponding to each interest point.
9. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program, when being executed by a processor, is adapted to carry out the steps of the user interest detection method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210012104.9A CN114036403B (en) | 2022-01-07 | 2022-01-07 | User interest detection method, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210012104.9A CN114036403B (en) | 2022-01-07 | 2022-01-07 | User interest detection method, device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114036403A CN114036403A (en) | 2022-02-11 |
| CN114036403B true CN114036403B (en) | 2022-03-25 |
Family
ID=80141328
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210012104.9A Active CN114036403B (en) | 2022-01-07 | 2022-01-07 | User interest detection method, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114036403B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116796076B (en) * | 2023-08-29 | 2023-11-03 | 中亿(深圳)信息科技有限公司 | Service recommendation method, device, equipment and storage medium |
| CN117056611B (en) * | 2023-10-11 | 2024-01-23 | 北京小糖科技有限责任公司 | New interest probing method, device, electronic equipment and storage medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102326144A (en) * | 2008-12-12 | 2012-01-18 | 阿迪吉欧有限责任公司 | The information that the usability interest worlds are confirmed is offered suggestions |
| CN104008184A (en) * | 2014-06-10 | 2014-08-27 | 百度在线网络技术（北京）有限公司 | Method and device for pushing information |
| CN104537115A (en) * | 2015-01-21 | 2015-04-22 | 北京字节跳动科技有限公司 | Method and device for exploring user interests |
| CN107229666A (en) * | 2016-12-30 | 2017-10-03 | 北京字节跳动科技有限公司 | A kind of interest heuristic approach and device based on commending system |
| CN108595461A (en) * | 2018-01-05 | 2018-09-28 | 武汉斗鱼网络科技有限公司 | Interest heuristic approach, storage medium, electronic equipment and system |
| CN108681919A (en) * | 2018-05-10 | 2018-10-19 | 苏州跃盟信息科技有限公司 | A kind of content delivery method and device |
| CN108804619A (en) * | 2018-05-31 | 2018-11-13 | 腾讯科技（深圳）有限公司 | Interest preference prediction technique, device, computer equipment and storage medium |
| CN110147481A (en) * | 2017-08-24 | 2019-08-20 | 腾讯科技（北京）有限公司 | Media content method for pushing, device and storage medium |
| US11106911B1 (en) * | 2018-06-13 | 2021-08-31 | Pointivo, Inc. | Image acquisition planning systems and methods used to generate information for structures of interest |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9852193B2 (en) * | 2009-08-10 | 2017-12-26 | Ebay Inc. | Probabilistic clustering of an item |
| CN106445969B (en) * | 2015-08-11 | 2019-03-05 | 北京字节跳动科技有限公司 | A kind of overall situation interest explores recommended method and device |
| US10719791B2 (en) * | 2017-05-30 | 2020-07-21 | Microsoft Technology Licensing, Llc | Topic-based place of interest discovery feed |
| CN108460101B (en) * | 2018-02-05 | 2019-05-28 | 山东师范大学 | Point of interest recommended method of the facing position social networks based on geographical location regularization |
| CN112579876A (en) * | 2019-09-30 | 2021-03-30 | 北京京东尚科信息技术有限公司 | Information pushing method, device and system based on user interest and storage medium |
| EP3828803A1 (en) * | 2019-11-26 | 2021-06-02 | Naver Corporation | Ambient point-of-interest recommendation using look-alike groups |
-
2022
- 2022-01-07 CN CN202210012104.9A patent/CN114036403B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102326144A (en) * | 2008-12-12 | 2012-01-18 | 阿迪吉欧有限责任公司 | The information that the usability interest worlds are confirmed is offered suggestions |
| CN104008184A (en) * | 2014-06-10 | 2014-08-27 | 百度在线网络技术（北京）有限公司 | Method and device for pushing information |
| CN104537115A (en) * | 2015-01-21 | 2015-04-22 | 北京字节跳动科技有限公司 | Method and device for exploring user interests |
| CN107229666A (en) * | 2016-12-30 | 2017-10-03 | 北京字节跳动科技有限公司 | A kind of interest heuristic approach and device based on commending system |
| CN110147481A (en) * | 2017-08-24 | 2019-08-20 | 腾讯科技（北京）有限公司 | Media content method for pushing, device and storage medium |
| CN108595461A (en) * | 2018-01-05 | 2018-09-28 | 武汉斗鱼网络科技有限公司 | Interest heuristic approach, storage medium, electronic equipment and system |
| CN108681919A (en) * | 2018-05-10 | 2018-10-19 | 苏州跃盟信息科技有限公司 | A kind of content delivery method and device |
| CN108804619A (en) * | 2018-05-31 | 2018-11-13 | 腾讯科技（深圳）有限公司 | Interest preference prediction technique, device, computer equipment and storage medium |
| US11106911B1 (en) * | 2018-06-13 | 2021-08-31 | Pointivo, Inc. | Image acquisition planning systems and methods used to generate information for structures of interest |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114036403A (en) | 2022-02-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114036403B (en) | User interest detection method, device and storage medium | |
| CN109190024B (en) | Information recommendation method and device, computer equipment and storage medium | |
| US20210295163A1 (en) | Neural architecture search | |
| CN111046152B (en) | Automatic FAQ question-answer pair construction method and device, computer equipment and storage medium | |
| US11157488B2 (en) | Reinforcement learning techniques to improve searching and/or to conserve computational and network resources | |
| CN109033408B (en) | Information pushing method and device, computer readable storage medium and electronic equipment | |
| CN108334575B (en) | Recommendation result sorting correction method and device and electronic equipment | |
| US9355095B2 (en) | Click noise characterization model | |
| EP4053716A1 (en) | Information recommendation method, apparatus, recommendation server and storage medium | |
| Paquet et al. | A hierarchical model for ordinal matrix factorization | |
| Hou et al. | Memory effect of the online user preference | |
| EP3688673A1 (en) | Neural architecture search | |
| CN110597965A (en) | Sentiment polarity analysis method and device of article, electronic equipment and storage medium | |
| US20220015657A1 (en) | Processing eeg data with twin neural networks | |
| Hu et al. | Individual popularity and activity in online social systems | |
| CN111988642B (en) | Method, device, server and storage medium for recommending videos | |
| CN113836388A (en) | Information recommendation method and device, server and storage medium | |
| CN116431912A (en) | User portrait pushing method and device | |
| CN114662008B (en) | Click position factor improvement-based CTR hot content calculation method and device | |
| CN113378037B (en) | Tariff configuration acquisition method and tariff configuration acquisition device | |
| CN109408725B (en) | Method and apparatus for determining user interest | |
| CN113901337A (en) | Application recommendation method, device, equipment and computer-readable storage medium | |
| Câmara et al. | RULK: A Framework for Representing User Knowledge in Search-as-Learning | |
| CN110347818B (en) | Word segmentation statistical method and device, electronic equipment and computer readable storage medium | |
| CN116361564A (en) | Super-parameter iteration training method, iteration training device and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |