A media recommendation method and device, and an electronic device

Abstract

The application discloses a media recommendation method and device and electronic equipment, and relates to the field of information technology, to solve the problem that the existing media recommendation method has poor mining effect on the stable preference attribute of a user, and the accuracy of the recommendation is poor. The method comprises the following steps: obtaining historical media viewing data of a user, wherein the historical media viewing data comprises media viewing data of N continuous historical periods; determining a preferred media label in the media viewing data of each historical period of the user in the N historical periods respectively; determining a target media label according to the number of occurrences of each media label preferred by the user in the media viewing data of the N historical periods; and recommending target media containing the target media label to the user according to the target media label. The embodiment of the application can mine the media label preferred by the user stably, so that the preferred media of the user can be more accurately recommended based on the media label.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and in particular to a media asset recommendation method and device and electronic equipment. BACKGROUND

[0002] With the development of network information, product recommendation based on user interest is widely used in various fields. At present, there are many recommendation schemes based on user long-term and short-term interest, such as time decay, preference propagation learning, user long-term and short-term click sequence fusion, knowledge graph, etc. However, these methods have poor mining effect on user stable preference attributes, resulting in poor accuracy of recommendation. SUMMARY

[0003] Embodiments of the present application provide a media asset recommendation method, device and electronic equipment to solve the problem that the existing media asset recommendation method has poor mining effect on user stable preference attributes, resulting in poor accuracy of recommendation.

[0004] In a first aspect, the embodiments of the present application provide a media asset recommendation method, comprising:

[0005] obtaining historical media asset viewing data of a user, wherein the historical media asset viewing data comprises media asset viewing data of N consecutive historical periods, and N is an integer greater than 1;

[0006] determining a preferred media asset tag in the media asset viewing data of each historical period of the N historical periods for the user;

[0007] determining a target media asset tag according to the number of occurrences of each media asset tag preferred by the user in the media asset viewing data of the N historical periods;

[0008] recommending a target media asset containing the target media asset tag to the user according to the target media asset tag.

[0009] Optionally, the media asset viewing data of the last K historical periods of the N historical periods is first type media asset viewing data, and the media asset viewing data of the N-K historical periods other than the last K historical periods of the N historical periods is second type media asset viewing data, and K is a positive integer less than N;

[0010] The determining of the preferred media asset tag in the media asset viewing data of each historical period of the N historical periods for the user comprises:

[0011] processing the media asset tags of the second type media asset viewing data by using a multi-head self-attention algorithm to determine N-K subsets of media asset tags preferred by the user, wherein each media asset tag in each subset of media asset tags in the N-K subsets of media asset tags is associated with a corresponding preference value;

[0012] processing the media content tags of the first type of media content watching data by using a collaborative filtering algorithm to determine K subsets of media content tags preferred by the user, wherein each of the K subsets of media content tags is associated with a corresponding preference value;

[0013] determining the target media content tag according to the number of occurrences of each media content tag preferred by the user in the media content watching data of the N historical time periods, comprising:

[0014] enhancing the preference values of the first media content tags in the first media content tag set, wherein the first media content tag set comprises the N-K subsets of media content tags and the K subsets of media content tags, and the number of occurrences of the first media content tag in the first media content tag set is greater than 1;

[0015] determining the target media content tag according to the preference values of each media content tag in the first media content tag set after the enhancement processing.

[0016] Optionally, the enhancing the preference values of the first media content tags in the first media content tag set comprises:

[0017] calculating the enhanced preference value of the first media content tag according to the preference values and the number of occurrences of the first media content tag in each subset of the first media content tag set, respectively.

[0018] Optionally, before the enhancing the preference values of the first media content tags in the first media content tag set, the method further comprises:

[0019] obtaining target behavior data of the user, wherein the target behavior data comprises at least one of subscription media content data and collection media content data;

[0020] determining the preference value of each media content tag in a second media content tag set, wherein the second media content tag set comprises the media content tags in the target behavior data;

[0021] determining the target media content tag according to the preference values of each media content tag in the first media content tag set after the enhancement processing, comprising:

[0022] adjusting the preference value of a second media content tag in the first media content tag set after the enhancement processing, wherein the second media content tag is the same media content tag in the first media content tag set and the second media content tag set, and the adjusted preference value of the second media content tag is a weighted value of the preference values of the second media content tag in the first media content tag set and the second media content tag set, respectively;

[0023] According to the adjusted preference values of the media content tags in the first media content tag set and the second media content tag set, a target media content tag is determined.

[0024] Optionally, the adjusted preference value of the second media content tag is a preference value after limit processing on the weighted value.

[0025] In a second aspect, the embodiments of the present application further provide a media content recommendation device, comprising:

[0026] A first obtaining module is configured to obtain historical media content viewing data of a user, wherein the historical media content viewing data comprises media content viewing data in N consecutive historical time periods, and N is an integer greater than 1;

[0027] A first determining module is configured to determine a media content tag preferred by the user in each of the N historical time periods according to the media content viewing data in each of the N historical time periods.

[0028] A second determining module is configured to determine a target media content tag according to the number of occurrences of each media content tag preferred by the user in the media content viewing data in the N historical time periods.

[0029] A recommendation module is configured to recommend a target media content containing the target media content tag to the user according to the target media content tag.

[0030] Optionally, the media content viewing data in the last K historical time periods of the N historical time periods is first type media content viewing data, and the media content viewing data in the N-K historical time periods other than the last K historical time periods of the N historical time periods is second type media content viewing data, and K is a positive integer less than N.

[0031] The first determining module comprises:

[0032] A first determining unit is configured to determine N-K media content tag subsets preferred by the user by processing the media content tags of the second type media content viewing data using a multi-head self-attention algorithm, wherein each media content tag in each of the N-K media content tag subsets is associated with a corresponding preference value.

[0033] A second determining unit is configured to determine K media content tag subsets preferred by the user by processing the media content tags of the first type media content viewing data using a collaborative filtering algorithm, wherein each media content tag in each of the K media content tag subsets is associated with a corresponding preference value.

[0034] The second determining module comprises:

[0035] The processing unit is configured to perform enhancement processing on the preference value of a first media tag in a first media tag set, wherein the first media tag set comprises the N-K media tag subsets and the K media tag subsets, and the first media tag appears in the first media tag set more than once.

[0036] The third determining unit is configured to determine a target media tag according to the preference values of the media tags in the first media tag set after the enhancement processing.

[0037] Optionally, the processing unit is configured to calculate an enhanced preference value of the first media tag according to the preference values and the appearance times of the first media tag in each subset of the first media tag set.

[0038] Optionally, the media recommendation apparatus further comprises:

[0039] The second obtaining module is configured to obtain target behavior data of a user, wherein the target behavior data comprises at least one of subscription media data and collection media data.

[0040] The third determining module is configured to determine a preference value of each media tag in a second media tag set, wherein the second media tag set comprises media tags in the target behavior data.

[0041] The third determining unit comprises:

[0042] The adjusting sub-unit is configured to adjust the preference value of a second media tag in the first media tag set after the enhancement processing, wherein the second media tag is a same media tag in the first media tag set and the second media tag set, and the adjusted preference value of the second media tag is a weighted value of the preference values of the second media tag in the first media tag set and the second media tag set.

[0043] The determining sub-unit is configured to determine a target media tag according to the preference values of the media tags in the first media tag set and the second media tag set after the adjustment.

[0044] Optionally, the adjusted preference value of the second media tag is a preference value after limit processing on the weighted value.

[0045] In a third aspect, an embodiment of the present application further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor performs the computer program to implement the steps in the media recommendation method as described above.

[0046] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the steps of the media recommendation method described above.

[0047] In the embodiments of the present application, the historical media viewing data of a user is acquired, wherein the historical media viewing data includes media viewing data of N consecutive historical time periods, and N is an integer greater than 1; preferred media tags of the user in the media viewing data of each historical time period of the N historical time periods are determined respectively; a target media tag is determined according to the number of occurrences of each media tag preferred by the user in the media viewing data of the N historical time periods; and a target media containing the target media tag is recommended to the user according to the target media tag. In this way, by analyzing the media viewing data of the user in each historical time period, the media tags preferred by the user in each historical time period are determined, and then the media tags stably preferred by the user are further locked based on the number of occurrences of each media tag in each historical time period, so that the preferred media of the user can be more accurately recommended based on the media tag. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required to be used in the description of the embodiments of the present application will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0049] Figure 1 is one of the flowcharts of the media recommendation method provided by the embodiments of the present application;

[0050] Figure 2 is a schematic diagram of feature enhancement using a meeting index provided by the embodiments of the present application;

[0051] Figure 3 is the second flowchart of the media recommendation method provided by the embodiments of the present application;

[0052] Figure 4 is a structural diagram of the media recommendation device provided by the embodiments of the present application;

[0053] Figure 5 is a structural diagram of the electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0054] Clearly, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0055] To make the embodiments of the present application clearer, the related technical knowledge involved in the embodiments of the present application will be introduced as follows:

[0056] With the help of network information development, product recommendation based on user interest has been widely applied in different fields. User interest modeling and personalized recommendation have been applied in various information services, but there are still many problems in product recommendation accuracy and long tail mining. There are many products of home broadband large screen, which are updated and replaced rapidly. When recommending, the long-term and short-term interests of users need to be considered, the long-term stable interest of users is found, and the current popular products are combined. For example, a user likes to watch a certain actor, but there is no new play for half a year. The user will have related brushing behavior and recent unrelated popular media behavior. The user will be recommended popular media in the conventional way, but the popular media does not meet the current preferences of all users. If the user has a long-term actor preference, recommending related media of the actor is more in line with the user's interest.

[0057] There are many existing recommendation schemes, including time decay, preference propagation learning, self-attention long-short-term fusion, knowledge graph, etc. The advantages and disadvantages of the prior art are as follows: when processing user long-short-term behavior, the user historical interest decays over time, the products with high recent interest intensity are preferentially recommended, and the historical behavior influence is reduced, which is not suitable for mining user stable preferences; the product recommendation algorithm based on user session interest divides the user behavior sequence more carefully, and assumes that the user's interest in a certain time period will not change much, which is suitable for stable product categories and user behavior intensive scenarios, while the user behavior of the home broadband large screen is long video watching, and there is no clear category between the media resources; the user long-short-term click sequence fusion method uses mechanisms such as gated recurrent mechanism to process user short-term behavior, and then calculates the user preference through a self-attention mechanism to obtain the user long-term click sequence, and then fuses the method of dot product attention processing user-click key-value pair sequence, which does not consider the connection between products, and does not reflect the hidden attribute connection between historical products and new hot products; the recommendation method based on knowledge graph, based on knowledge graph convolution network, uses entity neighborhood aggregation to aggregate the neighborhood entities to obtain the project feature vector representation of the recommended project, and the feature representation of the recommended project is calculated by inner product to obtain the corresponding feature similarity, which considers the internal connection of product features, and often recommends hot media resources with similar labels, but is easy to recommend new hot products, and cannot mine related products of user stable preference attributes, when new products that meet user interest are online, other products will be recommended to the user due to insufficient heat, and it takes a week or even longer time to accumulate heat.

[0058] In the embodiments of the present application, in order to recommend products that are more in line with the user's preferences at the moment, a media resource recommendation scheme based on user long-short-term interest is proposed, which can mine user stable preference features.

[0059] Referring to Figure 1 , Figure 1 is a flowchart of a media resource recommendation method provided by the embodiments of the present application, as shown in Figure 1 , comprising the following steps:

[0060] Step 101, obtaining historical media resource watching data of a user, wherein the historical media resource watching data includes media resource watching data of N consecutive historical time periods, and N is an integer greater than 1.

[0061] In order to mine the user's interest preferences, the historical media resource watching data of the user can be analyzed, so the historical media resource watching data of the user can be obtained first, including media resource information watched by the user, watching time, watching time length, etc.

[0062] Specifically, the media viewing data of the user in the past period of time can be acquired, such as the media viewing data of the user in the past year or the past three months. The acquired historical media viewing data can be divided into media viewing data of multiple stages according to time periods, such as media viewing data of N historical stages, specifically, the media viewing data of the user in the past year can be divided into media viewing data of 12 stages according to months.

[0063] Step 102, respectively determining the preferred media tags in the media viewing data of the user in each historical stage of the N historical stages.

[0064] In this step, for the media viewing data of the user in each historical stage, the preferred media tags of the user in each historical stage can be determined by analyzing the media tags, viewing time, viewing times, etc. of the viewed media, specifically, the preference values of the user for each media tag can be analyzed and determined using a related algorithm, so as to determine the preferred media tags of the user according to the preference values, such as determining the media tags with the preference values higher than a preset threshold as the preferred media tags of the user. The media tags can be tags representing characteristics or attributes of the media, such as the director, actor, type, era, suitable viewing age range, etc. of the media.

[0065] Step 103, determining the target media tag according to the number of occurrences of each media tag preferred by the user in the media viewing data of the N historical stages.

[0066] To mine the long-term stable preferences of the user, the number of occurrences of the media tags preferred by the user in each historical stage in the media viewing data of the N historical stages can be counted. When the user prefers the same media tag in multiple historical stages, the media tag occurs multiple times, and it can be determined that the preference degree of the user for the media tag is high, so the media tag can be determined as the target media tag of the stable preference of the user, or the preference value of the media tag can be recalculated in combination with the number of occurrences of the media tag and the preference value in each stage, and the recalculated preference value of the media tag is higher than the original preference value, and the media tag with the preference value in the front or the preference value greater than a certain value is selected as the target media tag of the stable preference of the user.

[0067] Optionally, the media viewing data of the K latest historical stages of the N historical stages is the first type of media viewing data, and the media viewing data of the N-K historical stages other than the K latest historical stages of the N historical stages is the second type of media viewing data, K is a positive integer less than N.

[0068] The step 102 comprises:

[0069] The multi-head self-attention algorithm is used to process the media asset tags of the second type of media asset watching data, to determine N-K media asset tag subsets preferred by the user, wherein each media asset tag in each media asset tag subset is associated with a corresponding preference value.

[0070] The collaborative filtering algorithm is used to process the media asset tags of the first type of media asset watching data, to determine K media asset tag subsets preferred by the user, wherein each media asset tag in each media asset tag subset is associated with a corresponding preference value.

[0071] The step 103 comprises:

[0072] The preference values of the first media asset tags in the first media asset tag set are enhanced, wherein the first media asset tag set comprises the N-K media asset tag subsets and the K media asset tag subsets, and the first media asset tag appears more than once in the first media asset tag set.

[0073] According to the preference values of the media asset tags in the first media asset tag set after the enhancement, a target media asset tag is determined.

[0074] In an implementation, the obtained media asset watching data of N historical periods can be divided into two types of media asset watching data according to different historical periods. Specifically, the media asset watching data of one or several historical periods close to the current time can be divided into the first type of media asset watching data, i.e., short-term media asset watching data, and the media asset watching data of the remaining historical periods far from the current time can be divided into the second type of media asset watching data, i.e., long-term media asset watching data. That is, the media asset watching data of the last K historical periods in the N historical periods is the long-term media asset watching data of the user, and the media asset watching data of the N-K historical periods except the last K historical periods in the N historical periods is the short-term media asset watching data of the user. In order to better represent the short-term watching behavior of the user, K can be 1.

[0075] For example, the obtained historical media asset watching data of the user can be divided into n stages according to time. Assuming that k represents the media asset watching data of a certain stage of the user, [n-1, n-2,..., n-k] is defined as the long-term and short-term media asset watching data of the user, wherein 1≤k≤n. When k=1, [n-1] belongs to the short-term media asset watching data of the user; when 1<k≤n, [n-2, n-3,..., n-k] belongs to the long-term media asset watching data of the user.

[0076] For the media asset watching data of each stage, a matrix can be used, wherein u represents the user, m represents the amount of media asset watched by the user, and 1≤i≤m.

[0077] Each media asset The attribute tags can include multiple types, such as director, actor, and other features. The other features are defined as features of the media asset other than the director and the actor, and the range includes but is not limited to media asset types such as comedy, costume, youth, city, animation, horror, and the like, suitable viewing age, era, and the like.

[0078] For each type of attribute tag of the media asset, an embedding can be used to represent it. Then, after embedding the three types of attribute tags of the media asset, the embedding of the three types of attribute tags is obtained. Combining the three types of attribute tags, the representation of the media asset viewing data of a user is obtained.

[0079]

[0080] Wherein, a i represents the current i-th media asset, F represents the combination of attribute tags, is the representation of the embedding of a certain attribute tag of the user.

[0081] Then, preference analysis can be performed on the media asset tags of the long-term media asset viewing data and the media asset tags of the short-term media asset viewing data of the user, respectively, to determine the long-term preferred media asset tags and the short-term preferred media asset tags of the user. Considering that the long-term media asset viewing data and the short-term media asset viewing data have different time spans and represent the preferences of the user at different times, different preference analysis algorithms can be used to determine the long-term and short-term preferred media asset tags of the user.

[0082] Specifically, for the second type of media asset viewing data, i.e., the long-term media asset viewing data of the user, the embedding of the media asset tags thereof can be constructed, and a multi-headed self-attention algorithm can be used to calculate the preference of the embedding of the media asset tags of the long-term media asset viewing data, to obtain the preference of the user for each media asset tag. The preference can be represented by a preference value, and the higher the preference value, the higher the preference of the user for the media asset tag. The media asset tag preferences of the user in different stages of media asset viewing data can be represented by multiple subsets, and the set of media asset tag preferences of the user is finally obtained. That is, the media asset viewing data of the N-K historical periods can be processed by preference analysis to obtain N-K media asset tag subsets, and each subset includes the preferred media asset tags of the user in the corresponding historical period. The N-K media asset tag subsets can be represented by .

[0083] For the first type of media asset watching data, i.e., short-term media asset watching data of the user, an Embedding of the media asset tag thereof can be constructed, and a preference of the user for each media asset tag can be calculated by using a collaborative filtering algorithm such as an ItemCF algorithm on the Embedding of the media asset tag of the short-term media asset watching data, and the preference can be specifically represented by using a preference value; wherein the media asset tag preferences of the user in different stages of media asset watching data can also be represented by using multiple subsets, and finally a set of media asset tag preferences of the user is obtained. That is, the media asset watching data of the last K historical periods can be processed by preference analysis to obtain K media asset tag subsets, and each subset includes the media asset tags preferred by the user in the corresponding historical period. The K media asset tag subsets can be respectively represented by using .

[0084] wherein t represents a set of media asset tags preferred by the user such as directors, actors, types, and the like, is a media asset tag feature representation preferred by the current user.

[0085] After determining the media asset tag set preferred by the user in the long and short term, media asset tags that appear multiple times in different subsets can be found to perform feature enhancement processing on the media asset tags that appear multiple times. The appearance of a certain media asset tag in different subsets indicates that the user has watched the media asset of the media asset tag in different historical periods, i.e., the user prefers the media asset tag, and thus the preference value of the media asset tag can be enhanced, i.e., the preference value of the media asset tag can be increased. For example, the media asset tag is a director, which appears in 3 subsets in the N-K media asset tag subsets and appears in 1 subset in the K media asset tag subsets, i.e., the director appears 4 times, and the preference value of the director can be enhanced, and the more the number of appearances, the more the preference value can be enhanced based on the original value.

[0086] After performing feature enhancement processing on the preference values of the media asset tags that appear multiple times, a target media asset tag can be determined according to the preference values of the media asset tags in the first media asset tag set after the enhancement processing, e.g., a certain number of media asset tags with high preference values can be determined as target media asset tags preferred by the user stably, or a media asset tag with a preference value greater than a predetermined threshold can be determined as a target media asset tag preferred by the user stably.

[0087] In this way, by using this embodiment, the media asset tags preferred by the user can be analyzed in combination with the long and short-term media asset watching behaviors of the user, and the media asset tags preferred by the user stably in the long term can be well mined by performing feature enhancement on the media asset tags preferred by the user in multiple periods.

[0088] Optionally, the enhancement processing on the preference value of the first media asset tag in the first media asset tag set comprises:

[0089] According to the preference values and the occurrence times of each subset of the first media tags, an enhanced preference value of the first media tag is calculated.

[0090] In an implementation, for a first media tag that occurs multiple times in the first media tag set, the enhanced preference value of the first media tag can be calculated by combining the preference values of the first media tag in each subset of media tags and the total occurrence times. For example, an index can be calculated according to the occurrence times of the first media tag, and then the sum of the preference values of the first media tag in each subset of media tags is multiplied by the index to obtain the enhanced preference value of the first media tag. Alternatively, the weight values of the first media tag in each subset of media tags can be determined according to the occurrence times of the first media tag, and then the preference values of the first media tag in each subset of media tags are multiplied by the corresponding weight values and then added together as the enhanced preference value of the first media tag. Of course, other calculation methods can also be used.

[0091] In this way, through this implementation, the media tags watched by the user in the long and short term can be reasonably enhanced in features, and the media tags that are long-term and stable preferences of the user can be further mined.

[0092] Optionally, before the preference values of the first media tags in the first media tag set are enhanced, the method further includes:

[0093] Obtaining target behavior data of the user, wherein the target behavior data includes at least one of subscription media data and collection media data;

[0094] Determining the preference values of each media tag in a second media tag set, wherein the second media tag set includes media tags in the target behavior data;

[0095] According to the preference values of each media tag in the first media tag set after the enhancement, determining a target media tag, includes:

[0096] Adjusting the preference value of a second media tag in the first media tag set after the enhancement, wherein the second media tag is a same media tag in the first media tag set and the second media tag set, and the adjusted preference value of the second media tag is a weighted value of the preference values of the second media tag in the first media tag set and the second media tag set, respectively;

[0097] According to the preference values of each media tag in the first media tag set and the second media tag set after the adjustment, determining a target media tag.

[0098] In an implementation, the specific behavior data of the user, such as the recently ordered media data and the collected media data, can also be combined to more accurately mine the stable preferred media tags of the user.

[0099] In this implementation, when the historical media viewing data of the user is obtained, the recently ordered media data and / or the collected media data of the user can also be obtained, and the media tags in the ordered media data and / or the collected media data can be determined based on the ordered media data and / or the collected media data, and the preference values of the user for the media tags can be statistically obtained. For example, if the user orders multiple media of a certain type, a higher preference value can be set for the media type. The set of the media tags preferred by the user in the target behavior data obtained through the statistics can be represented as

[0100] After the feature enhancement of the media tags that appear multiple times in the first media tag set, the preference values of the media tags can be further adjusted in combination with the media tags preferred in the ordering and collecting behaviors of the user. Specifically, when the first media tag set and the second media tag set of the ordered and collected media of the user have the same media tags, the preference values of the media tags in the two sets can be combined to perform a weighted processing on the preference values of the media tags, and the weighted preference values can be used as the new preference values of the media tags. Finally, the target media tags with higher preference values or exceeding a preset threshold can be determined according to the latest preference values of the media tags in the two sets.

[0101] In this way, by combining the ordering and collecting behaviors of the user and further adjusting the stable preferred media tags of the user, the user preferences mined can be more accurate, and the accuracy of the media recommended based on the user preferred media tags can be improved.

[0102] Optionally, the adjusted preference value of the second media tag is a preference value obtained by limiting the weighted value.

[0103] In an implementation, to avoid the adjusted preference value being too large or too small, the weighted preference value of the second media tag can be limited to a certain range. Specifically, a limiting function can be used to limit the weighted preference value of the second media tag, or when the weighted preference value of the second media tag exceeds a preset maximum value, the maximum value can be used to replace the weighted preference value, and when the weighted preference value of the second media tag is lower than a preset minimum value, the minimum value can be used to replace the weighted preference value.

[0104] The existing recommendation usually considers the recent behavior of the user or the viewing time length and the like, so that the new hot commodities are recommended more. In the embodiment of the application, it is considered that the long-term preference of the user is stable, and the meeting index is defined on the basis of considering and combining the long-term and short-term interests of the user and the label relationship between commodities to mine the long-term stable interest of the user.

[0105] Compared with the commonly used data conversion function, for example, the sigmoid function reduces the data range and the size relationship between data does not change. The meeting function set in the scheme can amplify the characteristics of the same preference of the user in multiple stages. The more the same preference appears in different stages, the larger the meeting index is and the size relationship between data will also change. If the same user watches more behaviors of the same label in multiple stages, the preference of the user for the attribute will increase exponentially. The meeting index is calculated as follows:

[0106] The input data includes: ti∈t, represents a media label of a certain attribute, represents a set of media labels of the long-term preference, represents a set of media labels of the short-term preference.

[0107] The meeting index is calculated as follows:

[0108]

[0109]

[0110] wherein, represents a characteristic of a certain media feature of a certain attribute in the continuous n stages (i.e. the set of media labels of the long-term and short-term user preferences) according to the above function, wherein k represents a certain media feature of a certain attribute of the current user preference, and x is a non-zero weight characteristic parameter, which can take a value between 0 and 1.

[0111] represents the adjustment of the user preference weight by the subscribed / collection media label, wherein ω 1i +ω 2i =1, 0≤ω 1i ≤1 and 0≤ω 2i ≤1, f i is a limit function whose maximum and minimum values are limited in a certain range.

[0112] Taking the subset "actors" of the set t as an example, it is assumed that the user's media viewing data in the past year is divided into 12 stages, the media viewing data in the past month is divided into short-term media viewing data, the media viewing data in the past 2-12 months is divided into long-term media viewing data of the user, the media subscribed / collected by the user in the past is divided into subscription / collection media data, and the preference feature expression of the user's media viewing data after Embedding expression and model calculation is L 1 , … L i , … L 11 , S, M, if the same actor appears multiple times in long-term and short-term media viewing data, that is, "meets" multiple times, the media viewing time of the actor is not the most recent one, but has been long-term attention of the user, and therefore the actor feature should be enhanced.

[0113] The following will be briefly described Figure 2 using the meeting index to enhance the feature:

[0114] It is assumed that x = 0.5, and actor a1 is watched in 5 stages of long-term and short-term media viewing data, that is, meets 4 times, so the result of actor a1 after fusion of the meeting index is 0.53; it is assumed that the viewing weight w1 = 0.4, and the subscription / collection weight w2 = 0.6, so the final result of actor a1 is 0.242, that is, the feature of actor a1 is enhanced from a score of 0.05 to 0.242.

[0115] Step 104, recommending target media containing the target media label to the user according to the target media label.

[0116] After determining the target media label of the stable preference of the user, the media of interest of the user can be recalled according to the target media label, and recommended to the user. Specifically, the target media whose media label contains the target media label can be recalled from the media viewed by all users, and the currently viewed media, subscribed media, etc. of the user can be added as multi-path recall, and the recall results are sorted to form the final recommendation results.

[0117] In combination with the above embodiments, the overall framework and data flow diagram of the embodiments of the present application can be as shown in Figure 3 .

[0118] The embodiments of the present application combine the 3-level long-term and short-term interests of the user, the long-term and short-term model fusion mines the preference labels of the user in different stages, and proposes a meeting amplification mechanism to effectively enhance the long-term preference. For example, if the user has a heavy brushing behavior, the related features will be strengthened, so as to more accurately mine the long-term stable interest preference of the user, and then more accurately recommend the media preferred by the user when the user is in a media drought or cold start of new media.

[0119] The method for recommending media asset provided in the embodiments of the present application comprises the following steps: obtaining historical media asset watching data of a user, wherein the historical media asset watching data comprises media asset watching data in N continuous historical time periods, N is an integer greater than 1; determining a media asset tag preferred by the user in the media asset watching data in each of the N historical time periods respectively; determining a target media asset tag according to the number of occurrences of each media asset tag preferred by the user in the media asset watching data in the N historical time periods; and recommending a target media asset containing the target media asset tag to the user according to the target media asset tag. In this way, by analyzing the media asset watching data of the user in each historical time period, the media asset tag preferred by the user in each historical time period is determined, and then the media asset tag preferred stably by the user is further locked based on the number of occurrences of each media asset tag in each historical time period, so that the media asset preferred by the user can be more accurately recommended to the user based on the media asset tag.

[0120] The embodiments of the present application further provide a device for recommending media asset. Referring to Figure 4 , Figure 4 is a structural diagram of the device for recommending media asset provided in the embodiments of the present application. Since the principle of solving problems of the device for recommending media asset is similar to that of the method for recommending media asset in the embodiments of the present application, the implementation of the device for recommending media asset can be referred to the implementation of the method, and the repeated parts will not be described herein.

[0121] As shown in Figure 4 , the device for recommending media asset 400 comprises:

[0122] A first obtaining module 401 is configured to obtain historical media asset watching data of a user, wherein the historical media asset watching data comprises media asset watching data in N continuous historical time periods, N is an integer greater than 1.

[0123] A first determining module 402 is configured to determine a media asset tag preferred by the user in the media asset watching data in each of the N historical time periods respectively.

[0124] A second determining module 403 is configured to determine a target media asset tag according to the number of occurrences of each media asset tag preferred by the user in the media asset watching data in the N historical time periods.

[0125] A recommending module 404 is configured to recommend a target media asset containing the target media asset tag to the user according to the target media asset tag.

[0126] Optionally, the media asset watching data in the last K historical time periods of the N historical time periods is first type media asset watching data, the media asset watching data in the N-K historical time periods except the last K historical time periods of the N historical time periods is second type media asset watching data, and K is a positive integer less than N.

[0127] The first determining module 402 comprises:

[0128] The first determining unit is configured to determine N-K media content label subsets preferred by the user by processing media content labels of the second type of media content watching data using a multi-head self-attention algorithm, wherein each media content label in each media content label subset is associated with a corresponding preference value.

[0129] The second determining unit is configured to determine K media content label subsets preferred by the user by processing media content labels of the first type of media content watching data using a collaborative filtering algorithm, wherein each media content label in each media content label subset is associated with a corresponding preference value.

[0130] The second determining module 403 includes:

[0131] The processing unit is configured to perform enhancement processing on the preference value of the first media content label in the first media content label set, wherein the first media content label set includes the N-K media content label subsets and the K media content label subsets, and the first media content label appears more than once in the first media content label set.

[0132] The third determining unit is configured to determine a target media content label according to the preference values of the media content labels in the first media content label set after the enhancement processing.

[0133] Optionally, the processing unit is configured to calculate an enhanced preference value of the first media content label according to the preference values and the number of appearances of the first media content label in each subset of the first media content label set.

[0134] Optionally, the media content recommendation apparatus 400 further includes:

[0135] The second obtaining module is configured to obtain target behavior data of the user, wherein the target behavior data includes at least one of subscription media content data and collection media content data.

[0136] The third determining module is configured to determine a preference value of each media content label in a second media content label set, wherein the second media content label set includes media content labels in the target behavior data.

[0137] The third determining unit includes:

[0138] The adjusting sub-unit is configured to adjust the preference value of a second media content label in the first media content label set after the enhancement processing, wherein the second media content label is a same media content label in the first media content label set and the second media content label set, and the adjusted preference value of the second media content label is a weighted value of the preference values of the second media content label in the first media content label set and the second media content label set.

[0139] determining sub-unit configured to determine a target media content label according to the adjusted preference values of each media content label in the first media content label set and the second media content label set.

[0140] Optionally, the adjusted preference value of the second media content label is a preference value after limit processing on the weighted value.

[0141] The media content recommendation apparatus 400 provided by the embodiments of the present application can execute the above-mentioned method embodiments, and has similar implementation principles and technical effects, which will not be repeated here.

[0142] The media content recommendation apparatus 400 provided by the embodiments of the present application acquires historical media content viewing data of a user, wherein the historical media content viewing data includes media content viewing data in N consecutive historical time periods, N is an integer greater than 1; determines a preferred media content label in the media content viewing data of each historical time period in the N historical time periods of the user respectively; determines a target media content label according to the occurrence times of each media content label preferred by the user in the media content viewing data in the N historical time periods; and recommends a target media content containing the target media content label to the user according to the target media content label. In this way, by analyzing the media content viewing data of the user in each historical time period, the media content label preferred by the user in each historical time period is determined, and then the media content label stably preferred by the user is further locked based on the occurrence times of each media content label in each historical time period, so that the preferred media content of the user can be more accurately recommended based on the media content label.

[0143] The embodiments of the present application further provide an electronic device. Since the principle of the electronic device to solve the problem is similar to the media content recommendation method in the embodiments of the present application, the implementation of the electronic device can be referred to the implementation of the method, and the repeated parts will not be repeated. As shown in Figure 5 The electronic device provided by the embodiments of the present application includes:

[0144] The processor 500 is configured to read the program in the memory 520 and execute the following process:

[0145] acquire historical media content viewing data of a user, wherein the historical media content viewing data includes media content viewing data in N consecutive historical time periods, N is an integer greater than 1;

[0146] determine a preferred media content label in the media content viewing data of each historical time period in the N historical time periods of the user respectively;

[0147] determine a target media content label according to the occurrence times of each media content label preferred by the user in the media content viewing data in the N historical time periods;

[0148] recommend a target media content containing the target media content label to the user according to the target media content label.

[0149] wherein, in Figure 5 the bus architecture can include any number of interconnected buses and bridges, specifically, various circuit links between the processor(s) 500 and the memory represented by the memory 520. The bus architecture can also link various other circuits such as peripheral devices, voltage stabilizers, and power management circuits, which are well known in the art, and thus, are not further described herein. The bus interface provides an interface. The processor 500 is responsible for managing the bus architecture and general processing, and the memory 520 can store data used by the processor 500 in performing operations.

[0150] Optionally, the processor 500 is further configured to read a program in the memory 520 and perform the following steps:

[0151] The media viewing data of the latest K historical time periods in the N historical time periods is first type media viewing data, the media viewing data of the N-K historical time periods other than the latest K historical time periods in the N historical time periods is second type media viewing data, and K is a positive integer less than N;

[0152] The media tags of the second type media viewing data are processed by using a multi-head self-attention algorithm to determine N-K media tag subsets preferred by the user, wherein each media tag in each media tag subset in the N-K media tag subsets is associated with a corresponding preference value;

[0153] The media tags of the first type media viewing data are processed by using a collaborative filtering algorithm to determine K media tag subsets preferred by the user, wherein each media tag in each media tag subset in the K media tag subsets is associated with a corresponding preference value;

[0154] The preference values of the first media tags in a first media tag set are enhanced, wherein the first media tag set includes the N-K media tag subsets and the K media tag subsets, and the first media tag appears more than once in the first media tag set;

[0155] A target media tag is determined according to the preference values of the media tags in the first media tag set after the enhancement.

[0156] Optionally, the processor 500 is further configured to read a program in the memory 520 and perform the following steps:

[0157] An enhanced preference value of the first media tag is calculated according to the preference values and the number of occurrences of the first media tag in each subset in the first media tag set.

[0158] Optionally, the processor 500 is further configured to read a program in the memory 520, and perform the following steps:

[0159] Obtain target behavior data of a user, wherein the target behavior data comprises at least one of subscription media data and collection media data;

[0160] Determine a preference value of each media tag in a second media tag set, wherein the second media tag set comprises media tags in the target behavior data;

[0161] Adjust the preference value of a second media tag in the first media tag set after the enhancement processing, wherein the second media tag is a same media tag in the first media tag set and the second media tag set, and the adjusted preference value of the second media tag is a weighted value of the preference values of the second media tag in the first media tag set and the second media tag set respectively;

[0162] Determine a target media tag according to the adjusted preference values of the media tags in the first media tag set and the second media tag set.

[0163] Optionally, the adjusted preference value of the second media tag is a preference value after limit processing on the weighted value.

[0164] The electronic device provided by the embodiment of the present application can execute the method embodiment described above, and has similar implementation principles and technical effects, which will not be repeated here.

[0165] In addition, the computer readable storage medium of the embodiment of the present application is used for storing a computer program, and the computer program can be executed by a processor to realize Figure 1 or Figure 3 the steps in the method embodiment shown in the method embodiment.

[0166] In several embodiments provided in the present application, it should be understood that the disclosed method and device can be implemented by other ways. For example, the device embodiment described above is only schematic, for example, the division of the unit is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units or components shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0167] In addition, each function unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically included separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of hardware plus software function unit.

[0168] The integrated unit realized in the form of software function unit can be stored in a computer readable storage medium. The software function unit is stored in a storage medium, including a plurality of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute part of the steps of the transceiving method described in each embodiment of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0169] The above is the preferred embodiment of the present application. It should be noted that, for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should also be considered as the protection scope of the present application.

Claims

1. A method of media recommendation, the method comprising: The method comprises: obtaining historical media viewing data of a user, wherein the historical media viewing data comprises media viewing data of N consecutive historical time periods, N is an integer greater than 1, media viewing data of K latest historical time periods in the N historical time periods is first type media viewing data, media viewing data of N-K historical time periods other than the K latest historical time periods in the N historical time periods is second type media viewing data, K is a positive integer less than N; processing media tags of the second type media viewing data by using a multi-head self-attention algorithm to determine N-K media tag subsets preferred by the user, wherein each media tag in each media tag subset in the N-K media tag subsets is associated with a corresponding preference value; processing media tags of the first type media viewing data by using a collaborative filtering algorithm to determine K media tag subsets preferred by the user, wherein each media tag in each media tag subset in the K media tag subsets is associated with a corresponding preference value; obtaining target behavior data of the user, wherein the target behavior data comprises at least one of subscription media data and collection media data; determining preference values of each media tag in a second media tag set, wherein the second media tag set comprises media tags in the target behavior data; performing enhancement processing on preference values of first media tags in a first media tag set, wherein the first media tag set comprises the N-K media tag subsets and the K media tag subsets, and the first media tags appear more than once in the first media tag set; adjusting preference values of second media tags in the first media tag set after the enhancement processing, wherein the second media tags are the same media tags in the first media tag set and the second media tag set, and the adjusted preference values of the second media tags are weighted values of the preference values of the second media tags in the first media tag set and the second media tag set respectively; determining target media tags according to the adjusted preference values of the media tags in the first media tag set and the second media tag set; recommending target media containing the target media tags to the user according to the target media tags.

2. The method of claim 1, wherein, The enhancement processing on the preference values of the first media tags in the first media tag set comprises: calculating enhanced preference values of the first media tags according to the preference values and the number of occurrences of the first media tags in each subset in the first media tag set.

3. The method of claim 1, wherein, The adjusted preference values of the second media tags are preference values obtained by limiting the weighted values.

4. A media recommendation apparatus characterized by comprising: The method comprises: The first obtaining module is configured to obtain historical media viewing data of a user, wherein the historical media viewing data comprises media viewing data of N continuous historical time periods, N is an integer greater than 1, media viewing data of K latest historical time periods in the N historical time periods is first type media viewing data, media viewing data of N-K historical time periods other than the K latest historical time periods in the N historical time periods is second type media viewing data, and K is a positive integer less than N; The first determining unit is configured to process media tags of the second type media viewing data by using a multi-head self-attention algorithm, and determine N-K media tag subsets preferred by the user, wherein each media tag in each media tag subset in the N-K media tag subsets is associated with a corresponding preference value; The second determining unit is configured to process media tags of the first type media viewing data by using a collaborative filtering algorithm, and determine K media tag subsets preferred by the user, wherein each media tag in each media tag subset in the K media tag subsets is associated with a corresponding preference value; The second obtaining module is configured to obtain target behavior data of the user, wherein the target behavior data comprises at least one of subscription media data and collection media data; The third determining module is configured to determine preference values of each media tag in a second media tag set, wherein the second media tag set comprises media tags in the target behavior data; The processing unit is configured to perform enhancement processing on preference values of first media tags in a first media tag set, wherein the first media tag set comprises the N-K media tag subsets and the K media tag subsets, and the first media tag appears more than once in the first media tag set; The adjusting subunit is configured to adjust preference values of second media tags in the first media tag set after the enhancement processing, wherein the second media tags are the same media tags in the first media tag set and the second media tag set, and the adjusted preference values of the second media tags are weighted values of the preference values of the second media tags in the first media tag set and the second media tag set respectively; The determining subunit is configured to determine target media tags according to the preference values of the media tags in the first media tag set and the second media tag set after the adjustment. The recommendation module is configured to recommend target media containing the target media tags to the user according to the target media tags.

5. The media recommendation apparatus of claim 4, wherein, The processing unit is configured to calculate enhanced preference values of the first media tags according to the preference values and the number of occurrences of the first media tags in each subset in the first media tag set.

6. An electronic device comprising: The memory, the processor, and the computer program stored in the memory and executable on the processor; the processor is configured to read the program in the memory to implement the steps in the media recommendation method according to any one of claims 1 to 3.

7. A computer readable storage medium for storing a computer program, characterized in that, The computer program is executed by the processor to implement the steps in the media recommendation method according to any one of claims 1 to 3.

Images