Content recommendation method, device and program product

Abstract

The invention discloses a content recommendation method and device and a program product, and relates to the technical field of data processing. According to the method, the content data and the behavior data of the same user on different devices are obtained, then the user portrait is constructed based on the behavior data, the content portrait is constructed based on the content data, and the user portrait represents preferences of the user for different types of content data in the preset time period. The content portrait represents the characteristics of the content data watched by the user in the preset time period, and finally, based on the content portrait and the user portrait, determining recommended contents on different devices. According to the method, the user portrait and the content portrait are established by integrating the watching data of the user on different devices, the preference of the user for different types of contents in different time periods is reflected, the behavior difference of the user on different devices is also considered, the interest change of the user can be more accurately captured and updated, and the user experience is improved. Therefore, the content recommendation accuracy is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a content recommendation method, device and program product. Background Art

[0002] With the rapid development of Internet technology, content recommendation systems have become an important tool for major Internet companies to improve user experience. Traditional recommendation methods are mainly based on the historical behavior of users on specific devices. They analyze the user's past behavior data to predict the user's future interests and needs, and then recommend content. However, this static content analysis method cannot integrate the user's viewing behavior on different devices, cannot capture the user's interest changes in content in real time, and cannot reflect the user's viewing preferences, thus affecting the accuracy of the recommendation.

[0003] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0004] The main purpose of this application is to provide a content recommendation method, device and program product, aiming to solve the technical problem of low accuracy of content recommendation.

[0005] To achieve the above objectives, the present application proposes a content recommendation method, which includes: Obtain content data and behavior data on different devices of the same user; Building a user profile based on the behavior data, wherein the user profile represents the user's preference for different types of content data within a preset time period; Building a content portrait based on the content data, wherein the content portrait represents the characteristics of the content data viewed by the user within a preset time period; Based on the content portrait and the user portrait, determine recommended content on different devices.

[0006] In one embodiment, the step of building a user profile based on the behavior data includes: Based on the behavior data, determining a content interaction action of the corresponding user within the preset time period; Based on the content interaction action, determining time sequence tags and corresponding interest scores of different types of the content data, wherein the time sequence tags represent a viewing time period of the content data; The user portrait is determined based on the content interaction action, the timing tag, and the interest score.

[0007] In one embodiment, the step of determining the time series labels and corresponding interest scores of different types of content data based on the content interaction action includes: Obtaining an associated weight value of the content interaction action; Based on the association weight values, interest scores corresponding to the different types of content data are obtained through weighted fusion.

[0008] In one embodiment, the step of constructing a content portrait based on the content data includes: extracting content features of the content data; Based on the type of the content data, determining type labels of different types of the content data; Determining a vector of the content data based on the content feature and the type label; The content portrait is determined based on the vector of the content data.

[0009] In one embodiment, the step of determining recommended content on different devices based on the content portrait and the user portrait includes: Get the usage time of the user's current device; Determining content to be recommended based on the user portrait and the usage time; Based on the content portrait, the recommended content is determined from the content to be recommended.

[0010] In one embodiment, the step of determining the content to be recommended based on the user portrait and the usage time includes: Matching a corresponding timing tag according to the usage time; If the interest score corresponding to the time sequence tag is greater than or equal to a preset threshold, the content data corresponding to the time sequence tag is determined as the content to be recommended.

[0011] In one embodiment, the step of determining the recommended content from the to-be-recommended content based on the content portrait includes: Based on the content portrait, determining the vector similarity of the content to be recommended, the vector similarity indicating the degree of similarity between the content to be recommended and the content data in the content portrait; The recommended content is determined based on the vector similarity.

[0012] In one embodiment, after the step of determining the recommended content on different devices based on the content portrait and the user portrait, the method further includes: Get the device type that the user is currently using; Determining a loading mode of the recommended content based on the device type; The recommended content is played based on the loading mode.

[0013] In addition, to achieve the above-mentioned purpose, the present application also proposes a content recommendation device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the content recommendation method described above.

[0014] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, wherein the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the content recommendation method described above are implemented.

[0015] The present application provides a content recommendation method, which obtains content data and behavior data on different devices of the same user, then constructs a user portrait based on the behavior data, and constructs a content portrait based on the content data, wherein the user portrait represents the user's preference for different types of content data within a preset time period, and the content portrait represents the characteristics of the content data watched by the user within the preset time period, and finally, based on the content portrait and the user portrait, determines the recommended content on different devices. The above method establishes user portraits and content portraits by integrating the viewing data of users on different devices, which not only reflects the user's preference for different types of content in different time periods, but also takes into account the differences in user behavior on different devices, and can more accurately capture and update changes in user interests, thereby improving the accuracy of content recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0018] Figure 1 A flowchart of the first embodiment of the method for recommending content of this application; Figure 2 A flowchart of the second embodiment of the method for recommending content of this application is provided; Figure 3 A flowchart of the third embodiment of the method for recommending content of this application is provided; Figure 4 A flowchart of the fourth embodiment of the method for recommending content of this application is provided; Figure 5 Schematic diagram of the device structure of the hardware operating environment involved in the content recommendation method in the embodiment of the present application.

[0019] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0020] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0021] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0022] With the rapid development of Internet technology, content recommendation systems have become an important tool for major Internet companies to improve user experience. Traditional recommendation methods are mainly based on the historical behavior of users on specific devices. They analyze the user's past behavior data to predict the user's future interests and needs, and then recommend content. However, this static content analysis method cannot integrate the user's viewing behavior on different devices, cannot capture the user's interest changes in content in real time, and cannot reflect the user's viewing preferences, thus affecting the accuracy of the recommendation.

[0023] In view of the above problems, the present application proposes a content recommendation method, which obtains content data on different devices and behavioral data generated by users watching content data, builds a user portrait based on the behavioral data, wherein the user portrait represents the user's preference for different types of content data within a preset time period, and at the same time, builds a content portrait based on the content data, wherein the content portrait represents the characteristics of the content data watched by the user within the preset time period, and finally, determines the recommended content on different devices based on the content portrait and the user portrait. The above method establishes user portraits and content portraits by integrating the viewing data of users on different devices, which not only reflects the user's preference for different types of content in different time periods, but also takes into account the differences in user behavior on different devices, and can more accurately capture and update changes in user interests, thereby improving the accuracy of content recommendations.

[0024] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, etc. The following takes the content recommendation system as an example to illustrate this embodiment and the following embodiments.

[0025] Based on this, the first embodiment proposed in this application provides a content recommendation method, referring to Figure 1 In this embodiment, the content recommendation method includes steps S10 to S40: Step S10, obtaining content data and behavior data on different devices of the same user.

[0026] It should be noted that content data refers to various types of information viewed by users on different devices, such as video, audio, graphics, etc., such as movies played on Internet TV, news articles browsed on smartphones, and educational videos watched on tablets. Behavioral data refers to various interactive information generated by users with content data on different devices, such as viewing time, click records, search history, comment interaction, etc., such as the time users watch movies on TV, the number of times they search for a certain type of news on their mobile phones, or the likes and comments on a certain educational video on a tablet.

[0027] For example, after obtaining user authorization, the user's behavioral data when using the device is recorded through the device's built-in sensors or applications, or in cooperation with a third-party data provider, the behavioral data of the same user on different platforms is integrated, and the collected data is uniformly stored and managed to provide a data foundation for the subsequent construction of user portraits and content portraits.

[0028] Step S20: constructing a user profile based on the behavior data, wherein the user profile represents the user's preference for different types of content data within a preset time period.

[0029] It should be noted that user portraits are used to describe the user's interests, preferences, and behavior patterns. For example, a user likes to watch movies on weekday evenings, but prefers to watch variety shows on weekends. User portraits can reflect the changes in user preferences for different types of content data in different time periods. For example, the above user prefers to watch lighthearted programs between 8 and 10 p.m., and prefers to watch news information in the morning.

[0030] Preprocess the collected behavioral data, including data cleaning, standardization, and feature extraction. Collect user behavioral data from different devices such as Internet TV, smart phones, tablets, etc., including but not limited to users' viewing history, search history, interactive behaviors such as likes, comments, sharing, etc., as well as viewing time and duration. Clean and standardize the collected behavioral data to remove duplicate, invalid or abnormal data to ensure data accuracy and consistency. Extract user viewing preference features in different time periods from the preprocessed behavioral data. These features may include the user's preference for different types of content such as movies, TV series, variety shows, news, etc., as well as viewing habits in different time periods such as weekday mornings and weekend evenings. Extract the temporal features of user behavior, such as the distribution of viewing time periods, changes in viewing frequency, etc., to reflect changes in user content preferences in different time periods.

[0031] Optionally, by analyzing the user's viewing records in different time periods, such as weekday mornings and weekend evenings, the user's viewing time and number of times in each time period are counted, and the user's content preferences in different time periods are inferred based on the time and number of times. For example, the user may prefer to watch movies or TV series in the evening, but may prefer short videos or audio content during commuting time.

[0032] Optionally, the viewing frequency of a user for a specific type of content within a period of time, such as a week or a month, is tracked, and a frequency change curve is plotted. The frequency change curve of a specific type of content reflects the changing trend of the user's interest. For example, when a user frequently watches a certain type of content, it may indicate that the user's interest in the content of this type is increasing.

[0033] Optionally, a time series model such as an autoregressive model, a moving average model or an autoregressive moving average model is used to model the viewing behavior of users in different time periods, capturing trends and periodic changes in the viewing behavior of users. Key features such as trend coefficient, cycle length, seasonal intensity, etc. are extracted from the time series model to reflect the temporal characteristics of the viewing behavior of users.

[0034] Optionally, a user profile is constructed based on the extracted user preference features. The user profile can also be combined with the user's demographic characteristics such as age, gender, region, etc. and interest propagation information in social networks to more accurately understand the user's needs and interests.

[0035] For example, during user registration or service use, basic demographic information of users, such as age, gender, and region, is collected through user information filling, device permission acquisition, etc., and user behavior data on social networks, such as following topics, participating in discussions, and sharing content, is obtained through API interfaces. The collected demographic characteristics and social network data are integrated with existing user behavior data to form a more comprehensive user data set. Using data mining and machine learning, features that can reflect user interests and preferences are extracted from the integrated user data set, such as user preference scores for different types of content, social influence indicators, and distribution of interest topics. Based on these features, a user portrait model is constructed to accurately describe user interests and preferences.

[0036] It should be noted that user portraits are not static, but will be dynamically updated as user behavior changes. As user behavior data continues to accumulate, user portraits can be updated and optimized regularly or in real time, and the accuracy and completeness of user portraits can be continuously improved by introducing new features, algorithms or behavior data.

[0037] For example, an incremental learning algorithm is designed to update the feature representation of users and content in real time. For new users, interest inference is performed based on their initial behavior and similar user groups. For example, the preferences and behavior patterns of these similar users are analyzed based on collaborative filtering algorithms or deep learning models to make preliminary inferences about the interests of new users.

[0038] Step S30, constructing a content portrait based on the content data, wherein the content portrait represents the characteristics of the content data viewed by the user within a preset time period.

[0039] It should be noted that content portraits are used to describe the characteristics, attributes and categories of content data. For example, the content portrait of a movie may include its genre such as action, comedy, director, actor, rating, release time and other characteristics.

[0040] Exemplarily, the content data is preprocessed to extract key features such as text descriptions, tags, visual and audio features, etc., and the content is automatically labeled and classified using multi-label classification models, knowledge graphs, and other methods. The extracted features and label information are fused to generate a unified content vector.

[0041] Optionally, step S30 includes steps S31 to S34: Step S31, extracting content features of the content data.

[0042] Optionally, the content data is input into a preset feature extraction model, and the model applies a corresponding processing algorithm based on the input data type such as text, image, audio, etc., to extract feature information representing the uniqueness of the content, including keywords in the text, objects in the image, melody features of the audio, etc.

[0043] For example, for text data, natural language processing techniques such as word segmentation, word vector representation, and topic models can be used to parse the text and extract keywords or themes of the text, such as news, comments, novels, etc. For image data, feature vectors extracted by convolutional neural networks can be used as vectors for content portraits, or visual features such as color distribution, texture, shape, and objects such as people and scenes in the image can be extracted. For audio data, MFCC (MelFrequency Cepstral Coefficients) extraction, audio classification algorithms, etc. can be used to parse the audio and extract features including audio rhythm, emotion, and sound type such as human voice, music, and ambient sound.

[0044] Step S32: determining type labels of different types of content data based on the type of the content data.

[0045] It should be noted that the type tag is a tag used to identify the category to which the content data belongs, such as "news", "movie", "music", etc.

[0046] Optionally, the type label can be determined by analyzing metadata of the content data, such as file name, extension, title, etc., or by analyzing the content of the content data, such as keywords of text content, visual features of image content, etc. The type label can also be determined by combining metadata and content analysis to improve accuracy.

[0047] Step S33: determining the vector of the content data based on the content feature and the type label.

[0048] Step S34, determining the content portrait based on the vector of the content data.

[0049] The extracted content features and the determined type labels are fused, and vector space models such as bag-of-words model, TF-IDF, word embedding, etc. are used to convert the fused information into mathematical vectors, and the obtained mathematical vectors are used as content portraits.

[0050] For example, suppose there is a piece of content data about a movie, and the features of the director, actors, genres such as action and science fiction have been extracted. The extracted content features such as director, actor names and genre labels such as "action" and "science fiction" are fused into a string or sequence. This string or sequence may contain multiple words, each of which corresponds to a feature or label in the content data. The fused string or sequence is input into the word embedding model to obtain the vector of each word. Then, these vectors are averaged, summed or other forms of aggregation operations are performed to obtain an overall content portrait vector.

[0051] Optionally, for new content, an initial portrait can be quickly constructed by calculating the feature similarity between the new content and the existing content. Specifically, the new content is first feature extracted to obtain the feature representation of its multimodal data such as text, images, and videos. Then, the feature representation of the new content is similar to the feature representation of the existing content. Based on the similarity calculation results, several existing contents that are most similar to the new content are selected, and the portraits of these existing contents are used as the initial portraits of the new content. As the new content is viewed and interacted by users, user feedback data on the new content is collected to continuously optimize the new content portrait to improve the accuracy of the content portrait.

[0052] Step S40: determining recommended content on different devices based on the content portrait and the user portrait.

[0053] By analyzing user behavior data on different devices and the characteristics of the content, we provide users with recommended content that best suits their interests and needs in different scenarios.

[0054] For example, suppose a user watches a science fiction movie on an Internet TV at home, and browses science fiction-related news and short videos on his mobile phone in the following days. Based on this information, the recommendation system can build a content portrait and a user portrait. Among them, the content portrait includes keywords, tags, visual features, etc. related to science fiction, and the user portrait indicates his strong interest in science fiction content and the habit of browsing science fiction-related content on different devices. When the user turns on the Internet TV or smartphone at home again, based on the large screen size of the TV and the characteristics of being suitable for watching long videos, the content recommendation system will recommend series or long content related to science fiction movies. Similarly, based on the small screen size of the mobile phone and the characteristics of being suitable for quick browsing, the content recommendation system will recommend short videos or news information related to science fiction.

[0055] In this embodiment, by collecting user behavior data from multiple devices in real time and integrating them in a unified manner, it is ensured that the user's behavior on different devices can be fully captured and analyzed, so that the content recommendation system can generate more representative user portraits and content portraits, thereby providing more accurate personalized recommendations.

[0056] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can refer to the above introduction, and will not be repeated later. Figure 2 , step S20 includes steps S21 to S23: Step S21: determining the content interaction actions of the corresponding user within the preset time period based on the behavior data.

[0057] It should be noted that content interaction refers to the interactive behavior between users and content, such as watching a video, clicking a link, searching for specific content, etc. Collect user behavior data on different devices in real time, and use natural language processing and machine learning technology to identify interactive actions between users and content, such as watching, searching, liking, etc. Classify the identified interactive actions, such as by content type such as movies, TV series, variety shows, etc. or by behavior type such as watching, searching, commenting, etc.

[0058] For example, the viewing behaviors of users on different devices are collected within one month, classified according to time periods, the types of content viewed in each time period and the corresponding viewing duration or frequency of the content are counted, and each interactive action is associated with a timestamp to form a time series.

[0059] Step S22: determining time series tags and corresponding interest scores of different types of content data based on the content interaction action, wherein the time series tags represent viewing time periods of the content data.

[0060] It should be noted that the interest score refers to the user's interest in the content calculated based on factors such as the frequency, duration, and degree of interaction between the user and the content.

[0061] Exemplarily, the time periods are first divided, for example, a day is divided into multiple time periods such as morning, morning, afternoon, evening, etc. According to the time when the user's content interaction action occurs, a corresponding time sequence label is generated for it, and the user's interest score for the content is calculated by weighted average, exponential weighting, etc., taking into account the frequency, duration, and degree of interaction between the user and the content.

[0062] Optionally, step S22 includes steps S221-S222: Step S221, obtaining the associated weight value of the content interaction action.

[0063] Step S222: obtaining interest scores corresponding to the different types of content data through weighted fusion based on the association weight values.

[0064] It should be noted that the association weight value refers to the importance of different types of content interaction actions such as watching, clicking, commenting, sharing, skipping, etc. in reflecting user interests. These weight values ​​are usually derived based on statistical analysis and empirical judgment, and are used to quantify the representativeness of different interaction actions to user interests. For example, the association weight of the watching action may be high because it directly reflects the user's interest in the content, while the association weight of the clicking action may be low because the click may be just the user's curiosity or mistaken operation, and does not necessarily mean that the user is really interested.

[0065] The user's content interaction action data on different devices are classified by type, and the associated weight value of each type is obtained. For each type of content data, a weighted calculation is performed based on the corresponding number or frequency of interaction actions and the associated weight value to obtain the interest score corresponding to the type of content data.

[0066] For example, assume that user Li Si watched content data multiple times through Internet TV, smartphones, and tablets in one day. Specific behaviors include watching a documentary on Internet TV in the morning, browsing 3 short videos on the mobile phone in the afternoon, and commenting and sharing a movie on the tablet in the evening. First, set the weight value of watching the entire content to 0.8, the weight value of clicking but not watching to 0.2, the weight value of commenting and sharing to 0.6, and you can also set some weight values ​​for negative feedback, such as the weight value of clicking to skip is -0.5, etc. After obtaining the associated weight values ​​of the content interaction actions, multiply the weight value of each interaction action by its corresponding number of interactions, and then perform the weighted summation of all interaction actions to obtain the interest score corresponding to the time series label. For example, the interest score corresponding to "Morning-Documentary" is 0.8 1=0.8; the interest score corresponding to "Afternoon-Short Video" is 0.2 3=0.6; the interest score corresponding to "Evening-Movie" is 0.6 1=0.6.

[0067] Step S23, determining the user portrait based on the content interaction action, the timing tag, and the interest score.

[0068] Exemplarily, the user's content interaction actions in different time periods, the corresponding time series tags and interest scores are integrated to form a data set that fully reflects the user's behavioral characteristics. By analyzing the integrated data, the user's preference patterns for different types of content in different time periods are identified. This includes the user's most active time period, the most favorite content type, and the content selection tendency in different situations. Based on the identified behavioral patterns, combined with other basic information of the user such as age, gender, region, etc., a multi-dimensional user portrait is constructed. This portrait not only describes the user's current interest status, but may also include the user's potential needs and possible future behavioral trends.

[0069] Optionally, the user profile is verified and optimized through continuous comparison and feedback with actual user behavior to ensure that it can accurately reflect the user's real needs and preferences. Exemplarily, a user feedback channel is established, such as online surveys, user comments, social media, etc., and the user profile is adjusted accordingly based on user feedback information.

[0070] In this embodiment, by analyzing the user's content interaction actions in different time periods, the user's interests and behavior patterns can be accurately captured, and the accuracy of the user portrait can be improved, thereby providing more sophisticated data support for the personalized recommendation system.

[0071] Based on the above embodiments of the present application, in the third embodiment of the present application, the same or similar contents as those in the above embodiments can be referred to the above introduction, and will not be described in detail later. Figure 3 , step S40 includes steps S41 to S43: Step S41, obtaining the usage time of the user's current device.

[0072] Step S42: Determine content to be recommended based on the user portrait and the usage time.

[0073] Optionally, the user's login time, operation time or active time period on the device is recorded to obtain the usage time. The user's interest preferences, historical behavior and other data in the user portrait are analyzed, and the user's content preferences in different time periods are identified in combination with the usage time. Based on these preferences, the recommended content that meets the user's current needs is screened from the content library.

[0074] Optionally, step S42 includes steps S421-S422: Step S421, matching a corresponding timing tag according to the usage time.

[0075] Exemplarily, the time the user is currently using the device is first recorded, and the time period to which the time belongs is determined. Then, the timing tags are extracted from the user portrait, and the current usage time is matched with the timing tags in the user portrait to find the tag that best matches the current time period.

[0076] Optionally, a day is divided into fixed time periods, such as morning, noon, and evening, and a time series label is assigned to each time period. Alternatively, the division of time periods is dynamically adjusted according to the user's daily behavior pattern, so that the time series label is more in line with the user's actual usage. For example, on weekdays, the time is divided into longer periods, such as morning, noon, and evening, while the frequency of users watching content will increase on weekends. At this time, the time is divided into shorter periods, in units of hours, to better meet the user's actual usage, thereby improving the accuracy and relevance of recommendations.

[0077] Step S422: If the interest score corresponding to the time sequence tag is greater than or equal to a preset threshold, determine the content data corresponding to the time sequence tag as the content to be recommended.

[0078] It should be noted that the preset threshold is used to determine whether the content data under a certain time series tag is worthy of being recommended to the user. After matching the time series tag of the current usage time, the interest score corresponding to the tag is found from the user portrait. Then, the interest score is compared with the preset threshold. If the interest score is greater than or equal to the threshold, it means that the user has a higher preference for the content type under the tag, so the content data corresponding to the tag is used as the content to be recommended.

[0079] Optionally, the preset threshold can be dynamically adjusted based on information such as the user's activity level, historical feedback, etc.

[0080] Exemplarily, define indicator data that can reflect the user's activity level, such as daily usage time, usage frequency, number of interactive behaviors, etc. Collect user feedback on recommended content, including positive feedback such as clicks, views, likes, etc., and negative feedback such as ignore, skip, complain, etc. Classify and quantify the collected feedback data to establish a prediction model. This model can be a simple linear regression model or a more complex machine learning model, such as a random forest, gradient boosting tree, etc. Use historical feedback data and historical activity data to train the prediction model so that it can predict the user's acceptance of recommended content at different activity levels. According to the user's current activity level, a dynamic preset threshold is calculated through the prediction model.

[0081] Optionally, the user's location information is obtained using a GPS (Global Positioning System) or IP (Internet Protocol) address, and the user's content preferences in different locations such as home, office, and public transportation are analyzed to determine the content to be recommended.

[0082] Exemplarily, the GPS location information or IP address of the user's device is obtained, and sensitive location data is obfuscated or encrypted for storage to protect user privacy. Use the map service API (Application Programming Interface) or IP address resolution service to convert the obtained GPS coordinates or IP address into specific geographic location information, such as city, region, and further analyze the environment where the user may be located, such as shopping mall environment, office environment, etc. Combined with the user's historical behavior data, analyze the user's content preferences in different locations and adjust the recommendation strategy. For example, if the user prefers to watch family dramas or variety shows at home, but prefers to watch news or industry information in the office. When the user is in a specific location, the content preferred by the user in that location is recommended first.

[0083] Optionally, environmental factors such as light and noise level are inferred through device sensor data, and the type of content to be recommended is adjusted according to the environmental factors, for example, subtitle content is recommended in a noisy environment.

[0084] Exemplarily, the built-in sensors of the device are used to collect light and noise data in real time, and the light intensity and noise level of the user's current environment are inferred based on the data of the light sensor and the noise sensor. For example, if the light intensity is very low, it may mean that the user is in a dark room or night environment. If the noise level is very high, it may mean that the user is in a noisy environment, such as public transportation, busy streets, etc. In a dimly lit environment, content suitable for watching in a dark room or at night, such as movies, nightly news, etc., can be recommended. At the same time, parameters such as screen brightness and contrast can be adjusted to provide a more comfortable viewing experience. In a noisy environment, content with subtitles can be recommended so that users can understand the content even if they cannot hear the sound clearly.

[0085] Step S43: Determine the recommended content from the content to be recommended based on the content portrait.

[0086] The content portrait is built based on information such as the content's text description, tags, visual and audio features, and the content to be recommended is further screened to select the content that best meets the user's current needs and interests.

[0087] Optionally, step S43 includes steps S431-S432: Step S431, based on the content portrait, determining the vector similarity of the content to be recommended, the vector similarity indicating the degree of similarity between the content to be recommended and the content data in the content portrait.

[0088] Step S432: determining the recommended content based on the vector similarity.

[0089] It should be noted that vector similarity is an indicator used to measure the proximity of two vectors in a multidimensional space, and is used to indicate the similarity between contents.

[0090] Extract the features of the content to be recommended, including text description, tags, visual and audio features, etc., convert the extracted features into vectors through the pre-trained model, and calculate the similarity between the vector of the content to be recommended and the content data vector in the content portrait. Among them, the vector similarity calculation is based on the distance measurement in multidimensional space, such as cosine similarity, Euclidean distance, etc. By comparing the similarity between the vector of the content to be recommended and the vector of the content portrait, the closeness between the content to be recommended and the content portrait can be evaluated, so as to determine whether it meets the user's preferences.

[0091] Optionally, the vector similarities of all the content to be recommended are sorted, and the recommended content is determined according to a preset similarity threshold or the top N of the sorting results. Alternatively, a similarity threshold is set, and when the vector similarity of the content to be recommended exceeds the threshold, it is determined to be the recommended content.

[0092] In this embodiment, the accuracy of personalized content recommendation is improved by integrating user portraits, device usage time and environmental factors. By matching the user's interest preferences in the current time period and dynamically adjusting the recommendation threshold, the content recommendation system can more accurately identify the user's needs at a specific moment, thereby providing content that better suits the user's taste. In addition, by further refining the recommendation strategy using GPS and sensor data, the content recommendation system can optimize the recommended content based on the user's actual location and environmental conditions, thereby enhancing the user experience. This method not only improves the accuracy of recommendations, but also enhances the adaptability and timeliness of recommendations, providing users with richer and more personalized services.

[0093] Based on the above embodiments of the present application, in the fourth embodiment of the present application, the same or similar contents as those in the above embodiments can be referred to the above introduction, and will not be described in detail later. Figure 4 After step S40, the content recommendation method further includes steps S50 to S70: Step S50: obtaining the device type of the device currently used by the user.

[0094] Step S60: determining a loading mode of the recommended content based on the device type.

[0095] Step S70: Play the recommended content based on the loading mode.

[0096] Optionally, the user agent string is used to obtain the type of the user's current device, where the user agent string contains the type, version, operating system and other information of the device. The content recommendation system can identify the device type by parsing this information. According to different device types, the video resolution and bit rate can be dynamically adjusted, and the optimal audio encoding format can be selected.

[0097] For example, for Internet TV, since the network environment is usually good and users expect high playback quality, the content recommendation system can choose to stream or preload the entire video content; while for smartphones, since the network environment may be unstable and users may pay more attention to the immediacy of the content, the content recommendation system can choose to progressively load or only load the key frames of the video. The content recommendation system obtains the data of the recommended content from the server according to the loading method corresponding to different device types, decodes it into video or audio signals, and displays it to the user through the device's audio and video output devices.

[0098] Optionally, for different device types, when playing recommended content, adjust the playback interface layout and interaction mode, for example, design an immersive interface layout for large-screen devices such as Internet TVs, adopt a full-screen playback mode, add high-definition picture quality options, and design subtitles and menus suitable for large-screen reading, etc., and optimize a compact layout for one-handed operation for small-screen devices such as smartphones. Alternatively, design a sliding and clicking-friendly interface for touch-screen devices, and set button navigation for devices controlled by remote controllers, etc.

[0099] In this embodiment, according to the characteristics of different devices, the content recommendation system can dynamically adjust the quality of video and audio to ensure that the content is presented in the most suitable way. In addition, the content recommendation system is also optimized for the interface layout and interaction mode of different devices, whether it is a large-screen or small-screen device, it can provide a customized playback experience. This method not only improves the efficiency and quality of content playback, but also enhances the interactivity and satisfaction of users, making content recommendations more intimate and efficient.

[0100] Exemplarily, in order to help understand the implementation process of the content recommendation method obtained by combining this embodiment with the above embodiments, the content recommendation system involved in the above embodiments is further described. Specifically, the content recommendation system includes: a user behavior data collection module, which is used to collect user behavior data from different devices such as OTT TVs, smart phones, and tablets, such as viewing history, search history, and interactive behavior. These data will be transmitted to the data storage module for processing and analysis. The data storage module is used to store the collected user behavior data and content feature data, support the rapid reading and writing of large-scale data, and ensure the efficiency of the system. The large model algorithm engine analyzes and models user behavior data based on deep learning and natural language processing technology. The recommendation result display module is used to display the recommended content generated by the large model to the user. The recommendation results are optimized according to the user's device and usage scenario to ensure that the user has a consistent experience on each device. The user feedback collection module is used to collect user feedback data on the recommendation results so as to adjust and optimize them in subsequent model training.

[0101] Among them, the user behavior data acquisition module and the data storage module are connected through a data interface to ensure real-time data transmission. The data storage module provides data support to the large model algorithm engine to ensure the efficiency of model training and real-time recommendation. The output of the large model algorithm engine is fed back to the user through the recommendation result display module to form a complete recommendation closed loop. The user feedback collection module is connected to the data storage module to store the user feedback data in the system for subsequent model optimization.

[0102] For example, when users use services on different devices, user behavior data is first collected and stored in the data storage module. Then the large model algorithm engine is called to generate recommendations and the recommendation results are displayed to the user. After that, user feedback is collected. If the user feedback information is valid, the recommendation strategy is adjusted according to the feedback information. If it is invalid, the current strategy is maintained.

[0103] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the content recommendation method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0104] The present application provides a content recommendation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the content recommendation method in the above-mentioned embodiment 1.

[0105] Reference below Figure 5 , which shows a schematic diagram of the structure of a content recommendation device suitable for implementing an embodiment of the present application. The content recommendation device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The content recommendation device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0106] like Figure 5As shown, the content recommendation device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the content recommendation device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the content recommendation device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a content recommendation device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0107] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0108] The content recommendation device provided by the present application adopts the content recommendation method in the above embodiment, which can solve the technical problem of low accuracy of content recommendation. Compared with the prior art, the beneficial effects of the content recommendation device provided by the present application are the same as the beneficial effects of the content recommendation method provided by the above embodiment, and other technical features in the content recommendation device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0109] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0110] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0111] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned content recommendation method when executed by a processor.

[0112] The computer program product provided by the present application can solve the technical problem of low accuracy of content recommendation. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the content recommendation method provided by the above embodiment, which will not be repeated here.

[0113] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A content recommendation method, characterized in that: The method comprises: Obtain content data and behavior data on different devices of the same user; Building a user profile based on the behavior data, wherein the user profile represents the user's preference for different types of content data within a preset time period; Building a content portrait based on the content data, wherein the content portrait represents the characteristics of the content data viewed by the user within a preset time period; Based on the content portrait and the user portrait, determine recommended content on different devices.

2. The content recommendation method according to claim 1, characterized in that: The step of constructing a user portrait based on the behavior data includes: Based on the behavior data, determining a content interaction action of the corresponding user within the preset time period; Based on the content interaction action, determining time sequence tags and corresponding interest scores of different types of the content data, wherein the time sequence tags represent a viewing time period of the content data; The user portrait is determined based on the content interaction action, the timing tag, and the interest score.

3. The content recommendation method according to claim 2, characterized in that: The step of determining the time series labels and corresponding interest scores of different types of content data based on the content interaction action comprises: Obtaining an associated weight value of the content interaction action; Based on the association weight values, interest scores corresponding to the different types of content data are obtained through weighted fusion.

4. The content recommendation method according to claim 1, characterized in that: The step of constructing a content portrait based on the content data includes: extracting content features of the content data; Based on the type of the content data, determining type labels of different types of the content data; Determining a vector of the content data based on the content feature and the type label; The content portrait is determined based on the vector of the content data.

5. The content recommendation method according to claim 2, characterized in that: The step of determining recommended content on different devices based on the content portrait and the user portrait includes: Get the usage time of the user's current device; Determining content to be recommended based on the user portrait and the usage time; Based on the content portrait, the recommended content is determined from the content to be recommended.

6. The content recommendation method according to claim 5, characterized in that: The step of determining the content to be recommended based on the user portrait and the usage time includes: Matching a corresponding timing tag according to the usage time; If the interest score corresponding to the time sequence tag is greater than or equal to a preset threshold, the content data corresponding to the time sequence tag is determined as the content to be recommended.

7. The content recommendation method according to claim 5, characterized in that: The step of determining the recommended content from the to-be-recommended content based on the content portrait comprises: Based on the content portrait, determining the vector similarity of the content to be recommended, the vector similarity indicating the degree of similarity between the content to be recommended and the content data in the content portrait; The recommended content is determined based on the vector similarity.

8. The content recommendation method according to claim 1, characterized in that: After the step of determining the recommended content on different devices based on the content portrait and the user portrait, the method further includes: Get the device type that the user is currently using; Determining a loading mode of the recommended content based on the device type; The recommended content is played based on the loading mode.

9. A content recommendation device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the content recommendation method according to any one of claims 1 to 8.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the content recommendation method according to any one of claims 1 to 8 are implemented.

Images