Method, system, storage medium and program product for dialogue based on cinematographic content

Abstract

The embodiment of the application provides a dialogue method and system based on film and television content, a storage medium and a program product. The embodiment of the application can comprehensively consider the long-term preference and short-term behavior of a user through a user portrait deeply fused with long-term and short-term interests, and provide more accurate content recommendation and chat interaction. The embodiment of the application also ensures that the system can be adjusted in real time through a flexible updating mode, dynamically adapts according to the recent activities and preferences of the user, and makes the dialogue scene more consistent with the current situation of the user. In addition, the embodiment of the application also improves the participation and satisfaction of the user in the chat process through a virtual character, greatly improving the user experience.

Description

Technical Field

[0001] This application relates to the technical fields of Internet digital home television and conversational recommendation systems, specifically to a dialogue method, system, storage medium, and program product based on film and television content. Background Technology

[0002] Personalized movie and TV chat is an innovative entertainment interaction technology that combines advanced artificial intelligence and big data analysis to provide users with customized movie and TV viewing experiences and in-depth interactive experiences.

[0003] In the digital and personalized era, film and television entertainment is no longer limited to passive viewing, but rather a process of active participation and communication. Personalized film and television chat services embody this trend. Through interaction with a large-scale AI model, users can obtain an intelligent, responsive, and emotionally rich companion. The core of this service lies in its personalization capabilities. The AI ​​system constructs a multi-dimensional user profile by analyzing the user's historical viewing records, rating preferences, search habits, and social interactions. Based on this, recommendation algorithms can accurately push film and television content that the user may be interested in, providing personalized satisfaction whether it's niche art films or popular TV series.

[0004] Currently, personalized movie and TV show companionship solutions still have some shortcomings. For example, although personalized recommendation algorithms are constantly being optimized, they may still fail to accurately capture users' true interests and preferences, resulting in inaccurate movie and TV show recommendations that do not meet users' expectations. Summary of the Invention

[0005] At least one embodiment of this application provides a dialogue method, system, storage medium, and program product based on film and television content to solve the problem that the recommended film and television content is not accurate enough.

[0006] To solve the above-mentioned technical problems, this application is implemented as follows:

[0007] In a first aspect, embodiments of this application provide a dialogue method based on film and television content, including:

[0008] Receive the current content entered by the first user in the current conversation;

[0009] The context information in the current dialogue is analyzed to determine the user intent of the first user, wherein the context information of the current dialogue includes the historical content and the current content of the current dialogue, and the user intent includes at least the intent to recommend movies and TV shows.

[0010] If the user intent of the first user is the movie / TV show recommendation intent, key information in the context information of the current dialogue is extracted, and based on the key information, at least one candidate movie / TV show is obtained from at least one data source and a candidate set is formed; wherein, the key information includes at least one of the following: movie / TV show character name, cast and crew name, movie / TV show title, movie / TV show category information, movie / TV show song name, and movie / TV show duration;

[0011] Based on the long-term and short-term fusion profile of the first user, the candidate films and television shows in the candidate set are sorted, and the candidate films and television shows are recommended to the first user in the sorted order.

[0012] Optionally, based on the long-term and short-term fusion profile of the first user, the candidate films and television shows in the candidate set are sorted, including:

[0013] Based on the popularity of the candidate films and television series, a first part of the candidate films and television series is selected from the candidate set.

[0014] Based on the matching degree between the feature vectors of the candidate films and television programs and the long-term and short-term fusion profiles of the first user, a second part of candidate films and television programs is selected from the first part of candidate films and television programs.

[0015] According to at least one of the preset business rules, operational strategies and preset scenario requirements, the second part of candidate films and television programs are screened and / or sorted to obtain the third part of candidate films and television programs.

[0016] The candidate films and television programs in the third part are displayed to the first user.

[0017] Optionally, the method further includes:

[0018] When displaying the candidate films and television shows from the third part of the candidate films and television shows to the first user, a brief introduction and reasons for recommending the candidate films and television shows are also further displayed.

[0019] Optionally, the method further includes:

[0020] Based on the registration information of the first user, an initial first user profile is created for the first user.

[0021] Collect the historical viewing behavior data of the first user, and extract the historical film and television sequence watched by the first user in chronological order; divide the historical film and television sequence according to the length of a preset time period to obtain a historical film and television set within multiple time periods; generate a viewing interest vector for each time period based on the feature vector of each film and television in the historical film and television set.

[0022] Based on the viewing interest vectors of different users in the same time period, the similarity of viewing interests between different users is calculated, and based on the viewing interest similarity, at least one other user is selected, wherein the viewing interest similarity between the other user and the first user is greater than a preset similarity threshold; based on the viewing interest similarity, the viewing interest vectors of the at least one other user in the same time period are weighted and summed to obtain the collaborative interest vector of the first user in the time period.

[0023] The first user's movie-watching interest vector and collaborative interest vector under the same time period are weighted and summed to obtain the first user's comprehensive interest vector under the same time period.

[0024] Obtain the comprehensive interest vector of the first user in each time period under the first time period and the comprehensive interest vector of the first user in each time period under the second time period; wherein, the length of the first time period is greater than the length of the second time period;

[0025] The first user's comprehensive interest vectors under all first time periods up to the current time are weighted and summed to obtain a first sum; the first user's comprehensive interest vectors under all second time periods up to the current time are weighted and summed to obtain a second sum; wherein, the comprehensive interest vectors closer to the current time have higher weights; the first sum and the second sum are weighted and summed to obtain the long-term and short-term fusion profile of the first user.

[0026] Optionally, based on the feature vector of each film and television program in the historical film and television collection, a viewing interest vector for each time period is generated, including:

[0027] The feature vectors of each film and television program in the historical film and television collection within the time period are weighted and summed to obtain the viewing interest vector of the first user in each time period. The weight of the feature vector of a film and television program is positively correlated with the ratio of the number of times the film and television program is viewed within the time period to the total number of times it is viewed within the time period. When the number of times the film and television program is viewed within the time period changes, the weight and the viewing interest vector of the first user in the time period are updated.

[0028] Optionally, the user intent may also include a conversational chat intent, and the method may further include:

[0029] If the user intent of the first user is the aforementioned chat intent, a search statement is generated based on the current content, and the search statement is used to search the knowledge base to obtain search information.

[0030] Based on the first virtual character selected by the first user, a first virtual character model of the first virtual character is constructed, the first virtual character model including the appearance features, personality features and language style features of the first virtual character; based on the first virtual character model, a corresponding first chat framework is generated;

[0031] Calculate the similarity between the retrieved information and each topic in the knowledge base of the first virtual character, select the target topic with the highest similarity, and combine the retrieved information with the target topic to generate chat reply information;

[0032] The chat reply information is displayed in the first chat frame.

[0033] Optionally, the method further includes:

[0034] If the user intent of the first user cannot be determined, a first prompt message for intent probing is constructed based on the long-term and short-term fusion profile of the first user and the context information of the current dialogue. The first prompt message is displayed in the current dialogue, and the step of receiving the current content input by the first user in the current dialogue is re-executed.

[0035] Secondly, embodiments of this application provide a dialogue system based on film and television content, including:

[0036] The receiving module is used to receive the current content entered by the first user in the current conversation;

[0037] The intent analysis module is used to analyze the context information in the current dialogue to determine the user intent of the first user, wherein the context information of the current dialogue includes the historical content and the current content of the current dialogue, and the user intent includes at least the intent to recommend movies and TV shows.

[0038] The search and recommendation module is used to extract key information from the context of the current dialogue when the user intent of the first user is the film and television recommendation intent, and to obtain at least one candidate film and television from at least one data source and form a candidate set based on the key information; wherein, the key information includes at least one of the following: film and television character names, cast and crew names, film and television drama titles, film and television category information, film and television song titles, and film and television duration; according to the long-term and short-term fusion profile of the first user, the candidate film and television in the candidate set are sorted, and the candidate film and television are recommended to the first user in the sorted order.

[0039] Optionally, the system further includes:

[0040] The search recommendation module is further configured to: select a first part of candidate films and television shows from the candidate set based on the popularity of the candidate films and television shows; select a second part of candidate films and television shows from the first part of candidate films and television shows based on the matching degree between the feature vectors of the candidate films and television shows and the long-term and short-term fusion profiles of the first user; filter and / or sort the second part of candidate films and television shows according to at least one of preset business rules, operation strategies and preset scenario requirements to obtain a third part of candidate films and television shows; and display the candidate films and television shows in the third part of candidate films and television shows to the first user.

[0041] Optionally, the search recommendation module is further configured to display a brief introduction and recommendation reasons for the candidate films and television programs when displaying them to the first user from the third part of the candidate films and television programs.

[0042] Optionally, the system further includes:

[0043] The image generation module is used for:

[0044] Based on the registration information of the first user, an initial first user profile is created for the first user.

[0045] Collect the historical viewing behavior data of the first user, and extract the historical film and television sequence watched by the first user in chronological order; divide the historical film and television sequence according to the length of a preset time period to obtain a historical film and television set within multiple time periods; generate a viewing interest vector for each time period based on the feature vector of each film and television in the historical film and television set.

[0046] Based on the viewing interest vectors of different users in the same time period, the similarity of viewing interests between different users is calculated, and based on the viewing interest similarity, at least one other user is selected, wherein the viewing interest similarity between the other user and the first user is greater than a preset similarity threshold; based on the viewing interest similarity, the viewing interest vectors of the at least one other user in the same time period are weighted and summed to obtain the collaborative interest vector of the first user in the time period.

[0047] The first user's movie-watching interest vector and collaborative interest vector under the same time period are weighted and summed to obtain the first user's comprehensive interest vector under the same time period.

[0048] Obtain the comprehensive interest vector of the first user in each time period under the first time period and the comprehensive interest vector of the first user in each time period under the second time period; wherein, the length of the first time period is greater than the length of the second time period;

[0049] The first user's comprehensive interest vectors under all first time periods up to the current time are weighted and summed to obtain a first sum; the first user's comprehensive interest vectors under all second time periods up to the current time are weighted and summed to obtain a second sum; wherein, the comprehensive interest vectors closer to the current time have higher weights; the first sum and the second sum are weighted and summed to obtain the long-term and short-term fusion profile of the first user.

[0050] Optionally, the profile generation module is further configured to perform a weighted summation of the feature vectors of each film and television program in the historical film and television collection within the time period to obtain the viewing interest vector of the first user in each time period; wherein, the weight of the feature vector of a film and television program is positively correlated with the ratio of the number of times the film and television program is viewed within the time period to the total number of times it is viewed within the time period, and when the number of times it is viewed within the time period changes, the weight and the viewing interest vector of the first user in the time period are updated.

[0051] Optionally, the user intent may also include a conversational chat intent, and the system may further include:

[0052] The dialogue module is used to generate a search statement based on the current content when the user intent of the first user is the desired chat intent, and to use the search statement to search the knowledge base to obtain search information; to construct a first virtual character model of the first virtual character selected by the first user, the first virtual character model including the appearance features, personality features and language style features of the first virtual character; to generate a corresponding first chat frame based on the first virtual character model; to calculate the similarity between the search information and each topic in the knowledge base of the first virtual character, to select the target topic with the highest similarity, to combine the search information with the target topic to generate chat reply information; and to display the chat reply information in the first chat frame.

[0053] Optionally, the system further includes:

[0054] The follow-up questioning module is used to construct a first prompt message for follow-up questioning based on the first user's long-term and short-term fusion profile and the context information of the current dialogue when the user's intention cannot be determined. The first prompt message is then displayed in the current dialogue, and the receiving module is re-triggered to execute the step of receiving the current content input by the first user in the current dialogue.

[0055] Thirdly, embodiments of this application provide a dialogue system based on film and television content, including: a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the method described in the first aspect.

[0056] Fourthly, embodiments of this application provide a computer-readable storage medium storing a program that, when executed by a processor, implements the steps of the method described in either the first or second aspect.

[0057] Fifthly, embodiments of this application provide a computer program product, including computer instructions that, when executed by a processor, implement the steps of the method described in either the first or second aspect.

[0058] Compared with existing technologies, the dialogue method, system, storage medium, and program products based on film and television content provided in this application can provide personalized film and television content recommendations for users, making the recommended content more in line with user expectations and improving user experience. This application, through deep integration of user profiles showing both long-term and short-term interests, can comprehensively consider users' long-term preferences and short-term behaviors, providing more accurate content recommendations and chat interactions. This application also ensures the system can adjust in real time through flexible update methods, dynamically adapting to users' recent activities and preferences. This flexibility makes the chat more suitable for the user's current situation. Furthermore, this application can create virtual characters that resonate more emotionally with users. This makes chatting not just information transmission, but can guide deeper emotional exchange and interaction, increasing user engagement and satisfaction. Compared with current film and television chat technologies, this application strengthens the interactivity between people and virtual characters, allowing users to feel more realistic and emotionally rich companionship while watching films and television shows. This is crucial for improving user stickiness and satisfaction. In summary, the system solution improves user experience through personalization, real-time feedback, and emotional resonance. Attached Figure Description

[0059] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0060] Figure 1 This is a flowchart of a dialogue method based on film and television content according to an embodiment of this application;

[0061] Figure 2This is a structural diagram of a dialogue system based on film and television content according to an embodiment of this application;

[0062] Figure 3 This is another structural diagram of a dialogue system based on film and television content, according to an embodiment of this application. Detailed Implementation

[0063] The terms "first," "second," etc., used in this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, without limiting the number of objects; for example, the first object can be one or more. Furthermore, the term "and / or" in this application describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0064] The term "instruction" in this application can be either a direct instruction (or explicit instruction) or an indirect instruction (or implicit instruction). A direct instruction can be understood as one in which the sender explicitly informs the receiver of specific information, the operation to be performed, or the requested result, etc., in the instruction sent. An indirect instruction can be understood as one in which the receiver determines the corresponding information based on the instruction sent by the sender, or makes a judgment and determines the operation to be performed or the requested result, etc., based on the judgment result.

[0065] As described in the background section, the accuracy of film and television content recommendations in the prior art is still insufficient. To solve at least one of the above problems, this application provides a dialogue method based on film and television content, which can provide personalized film and television content recommendations for users, making the recommended film and television content more in line with the user's expectations and improving the user experience.

[0066] Please refer to Figure 1 This application provides a dialogue method based on film and television content, which is applied to a dialogue system based on film and television content. Figure 2 A system architecture diagram of a dialogue system according to an embodiment of this application is provided. Figure 1 As shown, the method includes:

[0067] Step 11: Receive the current content entered by the first user in the current conversation.

[0068] Here, the first user can input the current content through the dialog box of the current conversation. The current content can be text, voice, image or video, etc., and this application embodiment does not specifically limit it.

[0069] Step 12: Analyze the context information in the current dialogue to determine the user intent of the first user. The context information of the current dialogue includes the historical content and the current content of the current dialogue. The user intent includes at least the intent to recommend movies and TV shows.

[0070] Here, the contextual information in the current dialogue includes the current content input by the first user and the historical content of the current dialogue. Specifically, the historical content of the current dialogue can include the historical content input by the first user during previous interactions and the responses generated by the dialogue system. This application embodiment can analyze the contextual information in the current dialogue using an intent analysis model to determine the first user's intent. The user intent includes at least a movie / TV recommendation intent, i.e., recommending movie / TV content to the user based on the contextual information of the current session. The user intent may also include a conversational chat intent, i.e., engaging in interactive chat with the user.

[0071] Step 13: If the user intent of the first user is the movie / TV show recommendation intent, extract key information from the context information of the current dialogue, and based on the key information, obtain at least one candidate movie / TV show from at least one data source and form a candidate set; wherein, the key information includes at least one of the following: movie / TV show character name, actor / crew member name, movie / TV show title, movie / TV show category information, movie / TV show song title, and movie / TV show duration.

[0072] Here, depending on the different classification rules, the film and television classification information has different classification results. For example, according to film and television type, the film and television classification information can include feature films, musicals, opera films, animated films, educational films, documentaries, dubbed films, etc.; according to film and television theme, the film and television classification information can include action films, romance films, science fiction films, horror films, historical films, etc.; according to film and television performance style, the film and television classification information can include animated films, musicals, opera films, etc.; according to film and television genre, the film and television classification information can include comedy, tragedy, tragicomedy, etc. This application embodiment does not specifically limit the classification rules, and film and television classification information obtained by one or more of the above classification rules can be used. In step 13, a search can be performed from one or more data sources that provide film and television content to find film and television content that matches one or more of the key information, which are then used as candidate films and television programs. In addition, when obtaining candidate films and television programs in step 13, the popularity of the candidate films and television programs is further obtained. The popularity is generated based on the number of times the candidate films and television programs are played within a preset time period; for example, the more times they are played, the higher the popularity.

[0073] Step 14: Based on the long-term and short-term fusion profile of the first user, sort the candidate films and television shows in the candidate set, and recommend the candidate films and television shows to the first user in the sorted order.

[0074] In step 14, this embodiment of the application may select a first part of candidate films and television programs from the candidate set based on factors such as the popularity of the candidate films and television programs. For example, candidate films and television programs with a popularity greater than a first popularity threshold may be selected as the first part of candidate films and television programs.

[0075] Then, based on the matching degree between the feature vectors of the candidate films and television programs and the long-term and short-term fusion profiles of the first user, a second part of candidate films and television programs is selected from the first part of candidate films and television programs. For example, in this embodiment of the application, a first feature vector of each candidate film and television program can be extracted from the first part of candidate films and television programs, the matching degree (e.g., similarity) between the first feature vector and the second feature vector corresponding to the long-term and short-term fusion profile of the first user can be calculated, and then candidate films and television programs with a matching degree greater than the first matching degree threshold are selected as the second part of candidate films and television programs. Here, the first feature vector of the candidate film and television program can be represented by the feature values ​​of the film and television content in multiple preset dimensions. The multiple preset dimensions may specifically include at least one of the following: film and television character names, cast and crew names, film and television drama titles, film and television classification information, film and television song titles, film and television duration, etc. The specific method of obtaining the second feature vector corresponding to the long-term and short-term fusion profile of the first user will be described below.

[0076] Then, according to at least one of the preset business rules, operational strategies, and preset scenario requirements, the second part of candidate films and television programs is screened and / or sorted to obtain the third part of candidate films and television programs; wherein, the business rules, operational strategies, and preset scenario requirements are usually used to indicate the film and television content that needs to be prioritized for recommendation.

[0077] Finally, the candidate films and television shows from the third part of the candidate film and television list are displayed to the first user. In addition, when displaying the candidate films and television shows from the third part of the candidate film and television list to the first user, a brief introduction and reasons for recommending the candidate films and television shows can also be displayed.

[0078] Through the above steps, this application embodiment can provide personalized film and television content recommendations for the first user based on the long-term and short-term fusion profile of the first user, making the recommended film and television content more in line with the user's expectations and improving the user experience.

[0079] For example, when an intention to recommend movies or TV shows is identified, the system further delves into key information, such as characters, show titles, tags, and channels. Simultaneously, it fully integrates the first user's historical preference information (a fusion of the first user's long-term and short-term profiles). This information is used in subsequent recommendation processes. Then, a multi-path recall stage is initiated, broadly collecting a set of potentially relevant recommendation content candidates from multiple data sources or recommendation strategies. For example, potential recommendations are recalled from databases related to the user's historical viewing records, popular content libraries, and recommendation resource libraries based on similar user groups. Next, a coarse sorting process is performed, using broad criteria such as content popularity and initial matching with the user's historical preferences to initially screen and sort the large number of recalled recommendations, eliminating obviously irrelevant or low-quality candidates and narrowing the scope. Then, a fine sorting stage is entered, employing more complex and precise algorithm models that comprehensively consider multiple factors, such as recent user preference trends, in-depth features of the content (e.g., plot type, actor influence ratings), and the accuracy of matching with the current input question, to meticulously optimize the ranking of the coarsely sorted recommendations. Next, at the strategy layer, the carefully sorted recommended content is finalized and adjusted based on business rules, operational strategies, and specific scenario requirements. For example, relevant content may be prioritized during specific promotional events, or different recommendation weights may be adjusted according to different user levels. Finally, the carefully selected and sorted recommended content is presented to users in an intuitive and user-friendly way, such as displaying recommended movies, TV shows, and channel information in lists or cards on the screen, along with brief introductions and reasons for recommendation, allowing users to quickly understand and select content.

[0080] The following explains how to obtain the second feature vector corresponding to the long-term and short-term fusion profile of the first user.

[0081] When a user initially registers, their historical viewing behavior data is scarce or incomplete. Therefore, an initial user profile can be created based on their registration information, serving as the long-term / short-term fusion profile, thus yielding the corresponding feature vector. The user profile includes multiple dimensions, typically the same or similar to the dimensions of the feature vectors for the film / television content. Specifically, it may include at least one of the following dimensions: film / television character names, cast and crew names, film / television title, film / television category information, film / television song titles, and film / television duration.

[0082] As the dialogue system operates, it can collect the first user's historical viewing behavior data, which includes at least one of the following: film and television browsing history, film and television search keywords, clicked films and television shows, film and television viewing duration, film and television comment content, film and television likes, and film and television dislikes. Then, the historical film and television sequence viewed by the first user is extracted in chronological order; the historical film and television sequence is divided according to a preset time period, resulting in multiple historical film and television sets within each time period. Based on the feature vector of each film and television show in the historical film and television set, a viewing interest vector for each time period is generated. For example, assuming a time period of one week, the historical film and television set viewed each week is obtained. Then, for each week's historical film and television set, a viewing interest vector for that week is generated. For example, the feature vectors of each film and television show viewed in that week are weighted and summed according to their respective weights to obtain the viewing interest vector for that week. The weight of each historical film and television show is determined based on the user's historical viewing behavior data. For example, weights are set based on film and television comment content (such as ratings), film and television likes, film and television dislikes, and film and television viewing duration. Generally, higher user ratings give higher weight to movies / TV shows that have been liked or watched for longer periods, while lower user ratings give lower weight to movies / TV shows that have been disliked or watched for shorter periods. Additionally, this embodiment can collect chat interaction data between the first user and the dialogue system, extracting movie / TV show information and the first user's emotional tendency information. This allows for the setting of weights for relevant movies / TV shows based on the historical viewing behavior data and the chat interaction data. For example, based on the chat interaction data, the degree of the first user's liking or disliking of a specific movie / TV show can be determined, and a corresponding weight can be set. Generally, movies / TV shows that the user likes more have higher weights, and movies / TV shows that the user dislikes more have lower weights.

[0083] When collecting user viewing behavior data and chat interaction data, this application embodiment can record various user behaviors in the film and television chat system in real time through the dialogue system, including but not limited to film browsing history, search keywords, clicked film and television recommendations, viewing time, comment content, likes / dislikes, etc. This behavioral data can intuitively reflect the user's interests and viewing habits. In addition, by analyzing the text content during the user's chat with the dialogue system, information such as film types, actors, directors, and plot elements mentioned by the user can be extracted, as well as the user's emotional tendency (such as positive, negative, or neutral) towards different topics. For example, by using sentiment analysis algorithms in natural language processing technology, it can be determined whether the user's evaluation of a certain film is positive or negative.

[0084] In addition, embodiments of this application can also preprocess the collected data through data preprocessing. Specifically, this includes:

[0085] Data cleaning: Removing noisy data, such as invalid characters, duplicate records, and incorrect data formats. For example, correcting common spelling errors in user input to ensure data accuracy and consistency.

[0086] Feature extraction: Extracting valuable features from raw data for subsequent updates to user profiles. For movie browsing history, features such as movie type, year, and region can be extracted; for chat text, features such as keywords and topics can be extracted.

[0087] Then, based on the viewing interest vectors of different users in the same time period, the viewing interest similarity between different users is calculated, and based on the viewing interest similarity, at least one other user is selected, wherein the viewing interest similarity between the other user and the first user is greater than a preset similarity threshold; based on the viewing interest similarity, the viewing interest vectors of the at least one other user in the same time period are weighted and summed to obtain the collaborative interest vector of the first user in the time period.

[0088] The first user's movie-watching interest vector and collaborative interest vector under the same time period are weighted and summed to obtain the first user's comprehensive interest vector under the same time period.

[0089] The above methods can be used to obtain the comprehensive interest vector of the first user within a certain time period. The long-short fusion profile of this application integrates comprehensive interest vectors from two different time periods. For example, using a month as the first time period and a week as the second time period, the comprehensive interest vectors of the first user in each time period within the first time period and the comprehensive interest vectors of the first user in each time period within the second time period are obtained according to the above methods; wherein the length of the first time period is greater than the length of the second time period.

[0090] Then, the comprehensive interest vectors of the first user under all first time periods up to the current time are weighted and summed to obtain a first sum; the comprehensive interest vectors of the first user under all second time periods up to the current time are weighted and summed to obtain a second sum; wherein, the comprehensive interest vectors more recent than the current time have higher weights. Then, the first sum and the second sum are weighted and summed to obtain the long-term and short-term fusion profile of the first user.

[0091] The following is a specific example to illustrate this.

[0092] This application embodiment can set various dimensions of the user profile, including but not limited to one or more:

[0093] (1) Interest and preference dimension: including the degree of preference for various film types (such as action, comedy, romance, science fiction, etc.), as well as the preference for specific actors, directors, and themes.

[0094] (2) Emotional Tendency Dimension: Emotional attitudes towards different types of films and film topics are divided into positive, negative and neutral.

[0095] (3) Viewing habits dimension: such as viewing time distribution (daytime / evening, weekday / weekend), viewing duration preference (feature film / short film), viewing frequency, etc.

[0096] When a user first uses the system, an initial profile is built based on the user's registration information (such as age, gender, and region) and a small amount of initial behavioral data (such as the type of movie viewed for the first time). For example, suppose a user selected an age range of 20-30 years old, a male gender, and a Beijing region during registration, and viewed an action movie for the first time. In this case, the initial profile can be set with a certain initial preference weight for the action movie genre, while the preference weight for other genres is relatively low.

[0097] Subsequent embodiments of this application can update the user profile based on user viewing behavior data and other information. Specifically:

[0098] Let I be the collection of movies and TV shows watched by user u. u ={i1,i2,...,i n For each film and television series j Extract its feature vector in Indicates film and television i j The value at the k-th feature dimension (e.g., film / TV genre, director, lead actor, etc.).

[0099] User u's movie viewing interest vector It can be obtained by weighted averaging the feature vectors of the movies and TV shows they have watched. The calculation formula is as follows:

[0100]

[0101] in, Is it film and television i j The weighting can be determined based on factors such as user ratings of movies and TV shows, and the number of times they are watched. in It is the user's view on film and television. j The rating, It is the user's viewing of movies and TV shows. j The number of times, α1 and β1 are weighting coefficients, which can be adjusted according to the actual situation.

[0102] Find the set S of users with similar interests to user u. uFor each user v∈S in the set u Calculate the interest similarity sim(u,v) between users u and v.

[0103] Similarity can be calculated using methods such as cosine similarity and Pearson correlation coefficient. Taking cosine similarity as an example, the calculation formula is as follows:

[0104]

[0105] Among them, V u and V v These are the interest vectors of users u and v (which can be content-based interest vectors or vectors represented in other ways), where · represents the vector dot product and |V| represents the magnitude of the vector.

[0106] Collaborative interest vector of user u based on collaborative filtering It can be obtained by weighted averaging of the interest vectors of similar users, and the calculation formula is as follows:

[0107]

[0108] Content-based user interest vectors and collaborative interest vectors based on collaborative filtering By fusing the data, we obtain the comprehensive interest vector V of user u. u For example, a simple weighted summation method can be used:

[0109]

[0110] Wherein, γ is the fusion coefficient, with a value range of [0,1], which can be adjusted according to the actual situation, and is used to balance the contributions of content-based and collaborative filtering-based methods in user interest modeling.

[0111] To balance short-term changes in user interests with long-term stable preferences, a weighted fusion approach is adopted. Below is the formula for fusing short-term and long-term interests, taking into account multiple dimensions such as time decay and the influence of different interests.

[0112] Let the long-term and short-term fused profile (comprehensive interest value) be I(t), which can be expressed as:

[0113]

[0114] in:

[0115] L k (t) represents the value of the comprehensive interest vector in the k-th long-term period at time t.

[0116] S m(t) represents the value of the comprehensive interest vector in the m-th short time period at time t.

[0117] w k and v m These are the weights of the comprehensive interest vectors for long and short time periods, which can be set according to the importance of different sources.

[0118] T k It is the most recent update time of the comprehensive interest vector for the k-th long-term period.

[0119] U m It is the most recent update time of the comprehensive interest vector in the m-th short time period.

[0120] λ k and μ m These are the decay factors of the comprehensive interest vectors for long and short time periods, which determine the degree of influence of time on interest.

[0121] α2 and β2 are weighting coefficients that satisfy α2 + β2 = 1, representing the relative importance of the comprehensive interest vector over long and short time periods.

[0122] K and M represent the number of integrated interest vectors for long and short time periods, respectively.

[0123] The long-term period and the short-term period here are two different time periods selected in advance, with the long-term period lasting longer than the short-term period.

[0124] Long-term periodic interest component: This is achieved by weighted summation of historical interests and application of a time decay function. This causes the influence of older interest values ​​on current overall interest to gradually weaken.

[0125] Short-term interest component: Similarly, short-term interest is processed through weighted summation, but a time decay function is also applied. Used to adjust short-term interests over time.

[0126] Weighting and Importance: By setting α² and β², the importance of short-term and long-term interests in the overall evaluation can be precisely controlled. The aforementioned short-term and long-term user profiles can more comprehensively consider the complex relationship between short-term and long-term interests and are suitable for scenarios requiring detailed analysis of user interests.

[0127] This application embodiment can also update the user's long-term and short-term integrated profile in real time. For example, when generating viewing interest vectors for each time period based on the feature vectors of each film and television show in the historical film and television set, the feature vectors of each film and television show in the historical film and television set within the time period can be weighted and summed to obtain the first user's viewing interest vectors for each time period; wherein, the weight of a film and television show's feature vector is positively correlated with the ratio of the number of times the film and television show was watched within the time period to the total number of times it was watched within the time period, and when the number of times it was watched within the time period changes, the weight and the first user's viewing interest vector for the time period are updated.

[0128] For example, suppose the first user has a preference for a certain film and television genre, T. i The initial preference weights are w i,0 The number of times a user views this type of film or television show within time t is n. i Given N(t) and N(t) being the total number of times movies and TV shows are viewed, the preference weight update formula for this type is:

[0129]

[0130] Among them, w i (t) represents the updated first user's view on film and television type T. i The preference weight is calculated using α3, which is an adjustment coefficient used to control the magnitude of the update and can be adjusted according to the actual situation. For example, if α3 = 0.2, and a user browses 5 action movies within a certain period of time, with a total of 20 movies viewed, and the initial action movie preference weight is 0.3, then the updated action movie preference weight will be...

[0131] The update methods and triggering mechanisms for the user long-term and short-term fusion profiles in this application embodiment are as follows:

[0132] (1) Real-time updates

[0133] For key user behaviors, such as instant search operations and clicking to watch videos, the system collects data in real time and updates user profiles accordingly. This ensures that the next interaction after these actions immediately reflects the user's latest interests and preferences.

[0134] (2) Regular updates

[0135] At regular time intervals (such as hourly or daily), the cumulative behavioral data of users during that period is aggregated and analyzed to update the user profile more comprehensively. For example, every morning, the user's movie-watching behavior, chat content, etc. from the previous day are comprehensively processed to update various dimensions of the user profile.

[0136] (3) Event-triggered update

[0137] When a user experiences significant events, such as watching a particular type of film for the first time, writing a lengthy review of a film, or giving an extreme rating (very high or very low), the system immediately triggers a profile update mechanism to capture significant changes in the user's interests more promptly.

[0138] Through the aforementioned real-time user profile update mechanism, the movie and TV chat system can more accurately understand users' dynamic needs and changing interests, thereby providing more personalized and precise services and recommendations, and improving user experience and satisfaction. At the same time, the continuously optimized profile model also helps the system better adapt to the behavioral patterns and preferences of different users, improving the system's intelligence level.

[0139] In this embodiment, if the user intent determined in step 12 is a chat intent, the dialogue system can first rewrite the current daily input by the user. The purpose of rewriting is to optimize the question expression to better meet the requirements of subsequent knowledge base retrieval and model processing, such as removing some vague, colloquial, or redundant expressions and converting them into more standardized query statements. Then, a knowledge base retrieval is performed, searching for information related to the rewritten question in a pre-built, large knowledge base. The knowledge base contains rich knowledge entries, such as frequently asked questions, historical dialogue records, and general knowledge documents. Through keyword matching, semantic similarity search, and other technical means, possible answer information is quickly located. At the same time, a virtual character model is constructed, and a chat response framework with character characteristics is generated based on the virtual character's personality, language style, knowledge background, and other factors. Finally, the information retrieved from the knowledge base is combined with the virtual character model to generate personalized, character-characterized chat responses that are closely related to the user's question, and these responses are displayed to the user, making the dialogue process more natural, vivid, and targeted.

[0140] Specifically, the user intent determined in step 12 is a conversational chat intent. At this time, a search statement can be generated based on the current content input by the first user in step 11, and the search statement can be used to search the knowledge base to obtain search information. Then, based on the first virtual character selected by the first user, a first virtual character model of the first virtual character is constructed. The first virtual character model includes the appearance features, personality features, and language style features of the first virtual character.

[0141] For example, the following formula can be used to construct a virtual character model:

[0142] V = f(A, S, L)

[0143] Where V represents the virtual character model, A represents appearance features, S represents personality features, L represents language style features, and f represents the construction function, which can be implemented using existing large models.

[0144] Then, based on the first virtual character model, a corresponding first chat frame is generated. Specifically, the appearance of the first chat frame can be generated based on at least one of the appearance features, personality features, and language style features of the first virtual character model. For example, the appearance of the first chat frame includes the appearance features of the first virtual character model, etc.

[0145] Each virtual character has its own corresponding knowledge base, which includes multiple topics related to the virtual character. These topics are generated based on the virtual character's personality traits and language style characteristics. This application embodiment calculates the similarity between the retrieved information and each topic in the first virtual character's knowledge base, selects the target topic with the highest similarity, and combines the retrieved information with the target topic to generate chat reply information, thereby generating chat reply information with personalized characteristics of the virtual character.

[0146] Here, cosine similarity can be used to calculate the similarity between the retrieved information and topics in the virtual character knowledge base. Let the vector representation of the retrieved information be T. u =(t u1 ,t u2 ,…,t uk ), where t ui Let represent the feature corresponding to the i-th token in the retrieved information, where i ranges from 1 to k. The vector representation of a topic in the knowledge base is T. v =(t v1 ,t v2 ,…,t vk ), where t vi Let represent the feature corresponding to the i-th token in the topic, where i ranges from 1 to k. If the number of tokens in the search information differs from the number of tokens in the topic, data preprocessing can be used to convert them into the same number of tokens. Thus, the cosine similarity Sim(T) between the search information and the topic... u ,T v The formula for calculating ) is:

[0147]

[0148] Where k represents the dimension of the topic vector, and T u ·T v Let |T| denote the dot product of two vectors. u | and | T v | represent vectors T respectivelyu and T v The magnitude of the cosine similarity. The closer the cosine similarity value is to 1, the more similar the two topics are.

[0149] In this embodiment of the application, if the user intent of the first user cannot be determined in step 12, or if the user intent is determined to be an unknown intent, a first prompt message for intent probing can be constructed based on the long-short fusion profile of the first user and the context information of the current dialogue. The first prompt message is displayed in the current dialogue, and step 11 is re-executed to further receive the current content input by the first user in the current dialogue in order to redetermine the user intent.

[0150] After the initial conversation with the first user, the dialogue system collects all data related to the user's actions on the recommended content (e.g., clicking to view or saving recommendations), the conversation itself (even when the intention is unclear), and data from interactions with the user when the intention is ambiguous. This data is then used to update the user profile. By analyzing the user's behavior and expressions during the conversation, the system further refines and improves the user's interests, preferences, and behavioral patterns. This provides more accurate data support for subsequent user interactions and personalized services, continuously optimizing the overall system's interaction experience and service quality, enabling the system to better adapt to and meet changing user needs.

[0151] In addition, in this embodiment of the application, the dialogue system can construct a chat model using the following formula:

[0152] C = g(D, U)

[0153] Where C represents the chat model, D represents film and television data, U represents user data, and g represents the construction function, which can be implemented using existing large-scale models.

[0154] As can be seen from the above, this application embodiment, by deeply integrating user profiles based on both long-term and short-term interests, can comprehensively consider users' long-term preferences and short-term behaviors, providing more accurate content recommendations and chat interactions. This application embodiment also ensures that the system can adjust in real time through flexible update methods, dynamically adapting to users' recent activities and preferences, making the dialogue scenario more consistent with the user's current situation. Furthermore, this application embodiment also enhances user engagement and satisfaction during chats through virtual characters, significantly improving the user experience.

[0155] The various methods of the embodiments of this application have been described above. Apparatus for implementing the above methods will now be provided.

[0156] Please refer to Figure 2 The dialogue system based on film and television content provided in this application includes:

[0157] The receiving module is used to receive the current content entered by the first user in the current conversation;

[0158] The intent analysis module is used to analyze the context information in the current dialogue to determine the user intent of the first user. The context information of the current dialogue includes the historical content and the current content of the current dialogue, and the user intent includes at least the intent to recommend movies and TV shows.

[0159] The search and recommendation module is used to extract key information from the context of the current dialogue when the user intent of the first user is the film and television recommendation intent, and to obtain at least one candidate film and television from at least one data source and form a candidate set based on the key information; wherein, the key information includes at least one of the following: film and television character names, cast and crew names, film and television drama titles, film and television category information, film and television song titles, and film and television duration; according to the long-term and short-term fusion profile of the first user, the candidate film and television in the candidate set are sorted, and the candidate film and television are recommended to the first user in the sorted order.

[0160] Optionally, the search recommendation module is further configured to: filter a first part of candidate films and television shows from the candidate set based on the popularity of the candidate films and television shows; filter a second part of candidate films and television shows from the first part of candidate films and television shows based on the matching degree between the feature vectors of the candidate films and television shows and the long-term and short-term fusion profiles of the first user; filter and / or sort the second part of candidate films and television shows according to at least one of preset business rules, operation strategies and preset scenario requirements to obtain a third part of candidate films and television shows; and display the candidate films and television shows in the third part of candidate films and television shows to the first user.

[0161] Optionally, the search recommendation module is further configured to display a brief introduction and recommendation reasons for the candidate films and television programs when displaying them to the first user from the third part of the candidate films and television programs.

[0162] Optionally, the system further includes:

[0163] The image generation module is used for:

[0164] Based on the registration information of the first user, an initial first user profile is created for the first user.

[0165] Collect the historical viewing behavior data of the first user, and extract the historical film and television sequence watched by the first user in chronological order; divide the historical film and television sequence according to the length of a preset time period to obtain a historical film and television set within multiple time periods; generate a viewing interest vector for each time period based on the feature vector of each film and television in the historical film and television set.

[0166] Based on the viewing interest vectors of different users in the same time period, the similarity of viewing interests between different users is calculated, and based on the viewing interest similarity, at least one other user is selected, wherein the viewing interest similarity between the other user and the first user is greater than a preset similarity threshold; based on the viewing interest similarity, the viewing interest vectors of the at least one other user in the same time period are weighted and summed to obtain the collaborative interest vector of the first user in the time period.

[0167] The first user's movie-watching interest vector and collaborative interest vector under the same time period are weighted and summed to obtain the first user's comprehensive interest vector under the same time period.

[0168] Obtain the comprehensive interest vector of the first user in each time period under the first time period and the comprehensive interest vector of the first user in each time period under the second time period; wherein, the length of the first time period is greater than the length of the second time period;

[0169] The first user's comprehensive interest vectors under all first time periods up to the current time are weighted and summed to obtain a first sum; the first user's comprehensive interest vectors under all second time periods up to the current time are weighted and summed to obtain a second sum; wherein, the comprehensive interest vectors closer to the current time have higher weights; the first sum and the second sum are weighted and summed to obtain the long-term and short-term fusion profile of the first user.

[0170] Optionally, the profile generation module is further configured to perform a weighted summation of the feature vectors of each film and television program in the historical film and television collection within the time period to obtain the viewing interest vector of the first user in each time period; wherein, the weight of the feature vector of a film and television program is positively correlated with the ratio of the number of times the film and television program is viewed within the time period to the total number of times it is viewed within the time period, and when the number of times it is viewed within the time period changes, the weight and the viewing interest vector of the first user in the time period are updated.

[0171] Optionally, the user intent may also include a conversational chat intent, and the system may further include:

[0172] The dialogue module is used to generate a search statement based on the current content when the user intent of the first user is the desired chat intent, and to use the search statement to search the knowledge base to obtain search information; to construct a first virtual character model of the first virtual character selected by the first user, the first virtual character model including the appearance features, personality features and language style features of the first virtual character; to generate a corresponding first chat frame based on the first virtual character model; to calculate the similarity between the search information and each topic in the knowledge base of the first virtual character, to select the target topic with the highest similarity, to combine the search information with the target topic to generate chat reply information; and to display the chat reply information in the first chat frame.

[0173] Optionally, the system further includes:

[0174] The follow-up questioning module is used to construct a first prompt message for follow-up questioning based on the first user's long-term and short-term fusion profile and the context information of the current dialogue when the user's intention cannot be determined. The first prompt message is then displayed in the current dialogue, and the receiving module is re-triggered to execute the step of receiving the current content input by the first user in the current dialogue.

[0175] It should be noted that the system in this embodiment is a device corresponding to the method described above. The implementation methods in each of the above embodiments are applicable to the embodiments of this system and can achieve the same technical effect. The system provided in this application embodiment can implement all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0176] Another embodiment of this application provides a dialogue system based on film and television content, such as... Figure 3 As shown, it includes a transceiver 1110, a processor 1100, a memory 1120, and a program or instructions stored in the memory 1120 and executable on the processor 1100; when the processor 1100 executes the program or instructions, it implements the various processes of the above-described terminal-side method embodiments and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0177] The transceiver 1110 is used to receive and send data under the control of the processor 1100.

[0178] Among them, Figure 3In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits of one or more processors represented by processor 1100 and memory represented by memory 1120 together. The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. Transceiver 1110 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. For different user equipment, user interface 1130 can also be an interface capable of connecting external or internal devices, including but not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0179] The processor 1100 is responsible for managing the bus architecture and general processing, and the memory 1120 can store the data used by the processor 1100 when performing operations.

[0180] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the above-described dialogue method embodiment based on video content, achieving the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may include read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0181] This application also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they implement the various processes of the above-described dialogue method embodiment based on film and television content and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0182] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0183] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0184] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A dialogue method based on film and television content, characterized in that, include: Receive the current content entered by the first user in the current conversation; The context information in the current dialogue is analyzed to determine the user intent of the first user, wherein the context information of the current dialogue includes the historical content and the current content of the current dialogue, and the user intent includes at least the intent to recommend movies and TV shows. If the user intent of the first user is the movie / TV show recommendation intent, key information in the context information of the current dialogue is extracted, and based on the key information, at least one candidate movie / TV show is obtained from at least one data source and a candidate set is formed; wherein, the key information includes at least one of the following: movie / TV show character name, cast and crew name, movie / TV show title, movie / TV show category information, movie / TV show song title, and movie / TV show duration; Based on the long-term and short-term fusion profile of the first user, the candidate films and television shows in the candidate set are sorted, and the candidate films and television shows are recommended to the first user in the sorted order.

2. The method according to claim 1, characterized in that, Based on the long-term and short-term fusion profile of the first user, the candidate films and television shows in the candidate set are sorted, including: Based on the popularity of the candidate films and television series, a first part of the candidate films and television series is selected from the candidate set. Based on the matching degree between the feature vectors of the candidate films and television programs and the long-term and short-term fusion profiles of the first user, a second part of candidate films and television programs is selected from the first part of candidate films and television programs. According to at least one of the preset business rules, operational strategies and preset scenario requirements, the second part of candidate films and television programs are screened and / or sorted to obtain the third part of candidate films and television programs. The candidate films and television programs in the third part are displayed to the first user.

3. The method according to claim 2, characterized in that, Also includes: When displaying the candidate films and television shows from the third part of the candidate films and television shows to the first user, a brief introduction and reasons for recommending the candidate films and television shows are also further displayed.

4. The method according to any one of claims 1 to 3, characterized in that, Also includes: Based on the registration information of the first user, an initial first user profile is created for the first user. Collect the historical viewing behavior data of the first user, and extract the historical film and television sequence watched by the first user in chronological order; divide the historical film and television sequence according to the length of a preset time period to obtain a historical film and television set within multiple time periods; generate a viewing interest vector for each time period based on the feature vector of each film and television in the historical film and television set. Based on the viewing interest vectors of different users in the same time period, the similarity of viewing interests between different users is calculated, and based on the viewing interest similarity, at least one other user is selected, wherein the viewing interest similarity between the other user and the first user is greater than a preset similarity threshold; based on the viewing interest similarity, the viewing interest vectors of the at least one other user in the same time period are weighted and summed to obtain the collaborative interest vector of the first user in the time period. The first user's movie-watching interest vector and collaborative interest vector under the same time period are weighted and summed to obtain the first user's comprehensive interest vector under the same time period. Obtain the comprehensive interest vector of the first user in each time period under the first time period and the comprehensive interest vector of the first user in each time period under the second time period; wherein, the length of the first time period is greater than the length of the second time period; The first user's comprehensive interest vectors under all first time periods up to the current time are weighted and summed to obtain a first sum; the first user's comprehensive interest vectors under all second time periods up to the current time are weighted and summed to obtain a second sum; wherein, the comprehensive interest vectors closer to the current time have higher weights; the first sum and the second sum are weighted and summed to obtain the long-term and short-term fusion profile of the first user.

5. The method according to claim 4, characterized in that, Based on the feature vector of each film and television program in the historical film and television collection, a viewing interest vector for each time period is generated, including: The feature vectors of each film and television program in the historical film and television collection within the time period are weighted and summed to obtain the viewing interest vector of the first user in each time period. The weight of the feature vector of a film and television program is positively correlated with the ratio of the number of times the film and television program is viewed within the time period to the total number of times it is viewed within the time period. When the number of times the film and television program is viewed within the time period changes, the weight and the viewing interest vector of the first user in the time period are updated.

6. The method according to claim 1, characterized in that, The user intent also includes a conversational chat intent, and the method further includes: If the user intent of the first user is the aforementioned chat intent, a search statement is generated based on the current content, and the search statement is used to search the knowledge base to obtain search information. Based on the first virtual character selected by the first user, a first virtual character model of the first virtual character is constructed, the first virtual character model including the appearance features, personality features and language style features of the first virtual character; based on the first virtual character model, a corresponding first chat framework is generated; Calculate the similarity between the retrieved information and each topic in the knowledge base of the first virtual character, select the target topic with the highest similarity, and combine the retrieved information with the target topic to generate chat reply information; The chat reply information is displayed in the first chat frame.

7. The method according to claim 1, characterized in that, Also includes: If the user intent of the first user cannot be determined, a first prompt message for intent probing is constructed based on the long-term and short-term fusion profile of the first user and the context information of the current dialogue. The first prompt message is displayed in the current dialogue, and the step of receiving the current content input by the first user in the current dialogue is re-executed.

8. A dialogue system based on film and television content, characterized in that, include: The receiving module is used to receive the current content entered by the first user in the current conversation; The intent analysis module is used to analyze the context information in the current dialogue to determine the user intent of the first user, wherein the context information of the current dialogue includes the historical content and the current content of the current dialogue, and the user intent includes at least the intent to recommend movies and TV shows. The search and recommendation module is used to extract key information from the context of the current dialogue when the user intent of the first user is the film and television recommendation intent, and to obtain at least one candidate film and television from at least one data source and form a candidate set based on the key information; wherein, the key information includes at least one of the following: film and television character names, cast and crew names, film and television drama titles, film and television category information, film and television song titles, and film and television duration; according to the long-term and short-term fusion profile of the first user, the candidate film and television in the candidate set are sorted, and the candidate film and television are recommended to the first user in the sorted order.

9. The system according to claim 8, characterized in that, The search recommendation module is further configured to: select a first part of candidate films and television shows from the candidate set based on the popularity of the candidate films and television shows; select a second part of candidate films and television shows from the first part of candidate films and television shows based on the matching degree between the feature vectors of the candidate films and television shows and the long-term and short-term fusion profiles of the first user; filter and / or sort the second part of candidate films and television shows according to at least one of preset business rules, operation strategies and preset scenario requirements to obtain a third part of candidate films and television shows; and display the candidate films and television shows in the third part of candidate films and television shows to the first user.

10. The system according to claim 9, characterized in that, The search recommendation module is also used to further display a brief introduction and recommendation reason for the candidate films and television programs when displaying them to the first user from the third part of the candidate films and television programs.

11. The system according to any one of claims 8 to 10, characterized in that, Also includes: The image generation module is used for: Based on the registration information of the first user, an initial first user profile is created for the first user. Collect the historical viewing behavior data of the first user, and extract the historical film and television sequence watched by the first user in chronological order; divide the historical film and television sequence according to the length of a preset time period to obtain a historical film and television set within multiple time periods; generate a viewing interest vector for each time period based on the feature vector of each film and television in the historical film and television set. Based on the viewing interest vectors of different users in the same time period, the similarity of viewing interests between different users is calculated, and based on the viewing interest similarity, at least one other user is selected, wherein the viewing interest similarity between the other user and the first user is greater than a preset similarity threshold; based on the viewing interest similarity, the viewing interest vectors of the at least one other user in the same time period are weighted and summed to obtain the collaborative interest vector of the first user in the time period. The first user's movie-watching interest vector and collaborative interest vector under the same time period are weighted and summed to obtain the first user's comprehensive interest vector under the same time period. Obtain the comprehensive interest vector of the first user in each time period under the first time period and the comprehensive interest vector of the first user in each time period under the second time period; wherein, the length of the first time period is greater than the length of the second time period; The first user's comprehensive interest vectors under all first time periods up to the current time are weighted and summed to obtain a first sum; the first user's comprehensive interest vectors under all second time periods up to the current time are weighted and summed to obtain a second sum; wherein, the comprehensive interest vectors closer to the current time have higher weights; the first sum and the second sum are weighted and summed to obtain the long-term and short-term fusion profile of the first user.

12. The system according to claim 11, characterized in that, The profile generation module is further configured to perform a weighted summation of the feature vectors of each film and television program in the historical film and television collection within the time period to obtain the viewing interest vector of the first user in each time period; wherein, the weight of the feature vector of a film and television program is positively correlated with the ratio of the number of times the film and television program is viewed within the time period to the total number of times it is viewed within the time period, and when the number of times it is viewed within the time period changes, the weight and the viewing interest vector of the first user in the time period are updated.

13. The system according to claim 8, characterized in that, The user intent also includes a conversational chat intent, and the system further includes: The dialogue module is used to generate a search statement based on the current content when the user intent of the first user is the desired chat intent, and to use the search statement to search the knowledge base to obtain search information; to construct a first virtual character model of the first virtual character selected by the first user, the first virtual character model including the appearance features, personality features and language style features of the first virtual character; to generate a corresponding first chat frame based on the first virtual character model; to calculate the similarity between the search information and each topic in the knowledge base of the first virtual character, to select the target topic with the highest similarity, to combine the search information with the target topic to generate chat reply information; and to display the chat reply information in the first chat frame.

14. The system according to claim 8, characterized in that, Also includes: The follow-up questioning module is used to construct a first prompt message for follow-up questioning based on the first user's long-term and short-term fusion profile and the context information of the current dialogue when the user's intention cannot be determined. The first prompt message is then displayed in the current dialogue, and the receiving module is re-triggered to execute the step of receiving the current content input by the first user in the current dialogue.

15. A dialogue system based on film and television content, characterized in that, include: Transceiver, processor, memory, and programs or instructions stored in the memory and executable on the processor; When the processor executes the program or instructions, it implements the steps of the method as described in any one of claims 1 to 7.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 7.

17. A computer program product, characterized in that, Includes computer instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 7.

Images