Program dynamic arrangement method, system, equipment and medium

Abstract

The invention is applicable to the technical field of computers, and provides a program dynamic arrangement method, which comprises the following steps of: acquiring hot event data from a social media platform and a news platform in real time, and acquiring watching behavior data of a user on a television platform to obtain watching preference characteristics; performing natural language processing on the hot event data to obtain a hot keyword, and performing similarity calculation on the hot keyword and metadata information of a program material database to generate a hot matching result; generating an initial population based on the watching preference features and the hotspot matching result, optimizing the initial population by adopting a genetic algorithm, and generating a program arrangement sequence; and the change conditions of hot events and user preferences are monitored in real time, and the program arrangement sequence is dynamically adjusted. According to the method and the device, the automation and intelligence of the whole process from hotspot information acquisition, program matching to program list generation and adjustment are realized, the workload and subjectivity of manual programming are reduced, and the efficiency and the quality of programming are improved.

Description

Technical Field

[0001] This application belongs to the field of computer technology, and in particular relates to a method, system, device and medium for dynamic program scheduling. Background Technology

[0002] With the rapid development of the Internet and digital media, viewers have placed higher demands on the real-time nature and diversity of television program content. Traditional television program scheduling mainly relies on human experience, which is inefficient and makes it difficult to respond in real time to social hot events or changes in audience preferences. Although the manual scheduling method can incorporate the professional experience of directors, it has significant drawbacks such as delayed response, low personalization, and insufficient diversity when faced with massive amounts of program content and a rapidly changing media environment.

[0003] In recent years, with the advancement of artificial intelligence and big data technologies, some video platforms and smart TV systems have begun to adopt recommendation algorithms to achieve personalized program content delivery. Specifically, existing program scheduling systems mainly adopt the following technical methods: (1) Rule-based static scheduling, which generates program lists according to preset schedules, program types, and fixed priorities, but usually lacks the ability to respond to real-time hot topics and dynamic user needs. (2) Recommendation algorithms based on user historical behavior, which match personalized program lists for users by analyzing their past behavior data, but still cannot achieve real-time adjustment of program lists and integration of hot topics. (3) Algorithm-based scheduling optimization, which generates program sequences by using optimization algorithms, which improves the diversity and matching degree of user interests to a certain extent, but the algorithm is complex and has poor real-time performance.

[0004] Therefore, the above-mentioned existing personalized program content estimation methods have improved the personalized recommendation capabilities to a certain extent, but still have the following defects: (1) Insufficient real-time performance: It is difficult to quickly capture sudden hot topics in social media and news, resulting in a delay in program schedule updates; (2) Limited semantic understanding: By relying on keyword matching, there is a lack of in-depth understanding of the context and semantic association of hot events; (3) Single arrangement strategy: It is impossible to dynamically balance the needs of the masses and niche groups, and the program is prone to repetition, leading to audience aesthetic fatigue; (4) High dependence on manual intervention: The arrangement rules and priorities still need to be set manually, and the level of automation and intelligence is limited; (5) Poor system generalization ability: It is less adaptable when facing texts in different fields or emerging program types; (6) Low processing efficiency: In particular, the computational resources of deep learning-based models are large, making it difficult to meet the real-time arrangement requirements; (7) Strong data dependence: Supervised learning methods require a large amount of labeled data, resulting in high manual costs and are not suitable for daily updated content. Summary of the Invention

[0005] This application provides a method, system, device, and medium for dynamic program scheduling, which can solve one of the above-mentioned problems in the prior art.

[0006] In a first aspect, embodiments of this application provide a method for dynamic program scheduling, including: Real-time data collection of trending events from social media and news platforms, as well as user viewing behavior data on TV platforms, to obtain viewing preference characteristics; Natural language processing is performed on the hot topic event data to obtain hot topic keywords. The similarity between the hot topic keywords and the metadata information of the program material database is calculated to generate hot topic matching results. Based on the viewing preference features and the hotspot matching results, an initial population is generated, and a genetic algorithm is used to optimize the initial population to generate the program arrangement order; The program schedule is dynamically adjusted based on real-time monitoring of trending events and changes in user preferences.

[0007] Furthermore, the real-time collection of trending event data from social media and news platforms includes: Using a distributed web crawler, hot word data is captured in real time from social media platforms, while news media API interfaces are called to obtain news events; The hot word data and the news events are associated and data is extracted to generate structured hot event data, which includes time, location and hot topic.

[0008] Furthermore, the collection of user viewing behavior data on the television platform to obtain viewing preference characteristics includes: By acquiring users' viewing history, search records, and channel switching behavior on TV platforms, and using machine learning algorithms to segment user groups, construct user profiles, and obtain viewing preference characteristics of different user groups, we can achieve the desired results.

[0009] Furthermore, the step of performing natural language processing on the hot topic event data to obtain hot topic keywords includes: Based on preset scoring parameters, the hot topic subjects of the hot topic event data are segmented in multiple levels to obtain core words; By combining the weighted fusion formula, the comprehensive score of each core word is calculated; The core words are sorted according to the comprehensive score, and the top N core words are selected as hot keywords.

[0010] Furthermore, the hot topic subject includes hot word data and event text of news events, and the preset scoring parameters include paragraph scoring threshold, sentence scoring threshold and keyword scoring coefficient; The process involves multi-level segmentation of the hot topic subjects in the hot topic event data based on preset scoring parameters to obtain core keywords, including: The hot topics are cleaned, and the cleaned hot words are marked as the first core words. The cleaned event text is segmented into multiple paragraphs. The semantic vector model is used to analyze the semantic relevance of each paragraph to the event text, generate paragraph scores, and determine the core paragraphs by combining the paragraph score thresholds. Each core paragraph is segmented into multiple sentences. An importance scoring model is used to analyze the importance of each sentence in the corresponding paragraph, generate a sentence score, and determine the core sentences by combining the sentence score threshold. The core sentence is segmented into multiple words. A word score is generated based on the paragraph score of the paragraph in which the word is located, the sentence importance score of the sentence in which the word is located, and the comprehensive score of the word. The second core word is determined by combining the keyword score coefficient.

[0011] Furthermore, the program material database includes multiple program materials, each of which is equipped with metadata information, including program type, theme content, broadcast duration, production time, and copyright information; The step of calculating the similarity between the hot keywords and the metadata information in the program material database to generate hotspot matching results includes: The semantic similarity between the hot keywords and the program type and theme content of the metadata information is calculated to obtain preliminary program materials; Determine the emotional tendency of trending events, and adjust the preliminary program materials based on the emotional tendency to obtain related program materials; The BERT model is used to perform entity recognition on the hotspot event data to obtain hotspot entities; The associated program materials are linked with the hotspot entities to construct an event-program association graph. In the event-program association graph, one hotspot entity corresponds to one or more program materials, generating hotspot matching results.

[0012] Furthermore, the process of generating an initial population based on the viewing preference features and the hotspot matching results, and optimizing the initial population using a genetic algorithm to generate the program arrangement order, includes: An initial population is generated based on the viewing preference features and the hotspot matching results; In the initial population, multiple programs are sequentially encoded into gene sequences, each of which includes a program ID, a playback time period, and a playback priority. Using user interest satisfaction, hot topic relevance, and content diversity as optimization objectives, a fitness function is constructed. Through selection, crossover, and mutation operations, the initial population is iteratively evolved until a preset iteration condition is met, resulting in an optimized program arrangement order.

[0013] Secondly, embodiments of this application provide a program dynamic scheduling system, including: Data acquisition module: used to collect real-time data on trending events from social media and news platforms, and at the same time collect user viewing behavior data on TV platforms to obtain viewing preference characteristics; Semantic analysis module: used to perform natural language processing on the hot event data to obtain hot keywords, calculate the similarity between the hot keywords and the metadata information of the program material database, and generate hot matching results; Genetic algorithm optimization module: used to generate an initial population based on the viewing preference features and the hotspot matching results, and to optimize the initial population using a genetic algorithm to generate the program arrangement order; Dynamic adjustment module: Used to monitor changes in trending events and user preferences in real time, and dynamically adjust the program arrangement order.

[0014] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described dynamic program scheduling method.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium, including a computer program stored in the computer-readable storage medium, wherein the computer program, when executed by a processor, implements the above-described dynamic program scheduling method.

[0016] Compared with the prior art, the embodiments of this application have at least one of the following advantages: (1) By using real-time hot spot capture and semantic analysis technology, the data of real-time hot events and program material database can be accurately semantically matched, which can closely link hot events with program content, thereby improving content relevance, ensuring that the program list can reflect the current hot topics in a timely manner, making the program list more timely and attractive, attracting audience attention and improving the timeliness and relevance of program content, so as to solve the technical problem that the existing technology is not sensitive enough to hot events and cannot meet the audience's attention to hot content in a timely manner.

[0017] (2) By optimizing the program order through a multi-objective genetic algorithm, taking into account user preferences and content diversity, it overcomes the limitations of existing technologies in balancing user interests and content diversity, effectively avoids the problem of homogenized recommendations, improves the viewing experience and satisfaction of the audience, reduces human intervention, improves the efficiency and accuracy of program arrangement, and adapts to the needs of multiple scenarios. (3) By performing natural language processing on hot event data, the annotation efficiency is improved, the annotation cost is reduced, the adaptability to new information is enhanced, the hot information updated daily is effectively processed, the response speed is faster, the real-time processing needs are met, the reliance on manually written rules by domain experts is reduced, the scalability and generalization ability are improved, the keyword annotation of the text can better adapt to different fields and contexts, the reliance on domain expert knowledge is reduced, and in this process, by combining word frequency, relation attributes and other elements, a multi-level segmentation method is adopted, which can more accurately extract keywords in long texts, meet the text analysis and information retrieval needs of professional fields, improve the accuracy and precision of sentence segmentation strategy, optimize the segmentation process, avoid the problem of overly fine word segmentation and special words that cannot be segmented, and improve the accuracy and effectiveness of keyword extraction.

[0018] In summary, the program dynamic scheduling method disclosed in this application achieves full automation and intelligence from hot topic information acquisition and program matching to program list generation and adjustment, reducing the workload and subjectivity of manual scheduling and improving the efficiency and quality of program scheduling. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a dynamic program scheduling method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a program dynamic scheduling system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0021] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0022] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0023] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0024] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0025] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0027] Please see Figure 1 As shown, the present invention is a method for dynamic program scheduling, comprising the following steps: S100 collects real-time data on trending events from social media and news platforms, while also collecting user viewing behavior data on TV platforms to obtain viewing preference characteristics. In some embodiments, the real-time collection of trending event data from social media and news platforms includes: Using a distributed web crawler, hot word data is captured in real time from social media platforms, while news media API interfaces are called to obtain news events; The hot word data and the news events are associated and data is extracted to generate structured hot event data, which includes time, location and hot topic.

[0028] In this embodiment, multiple web crawler programs are deployed to crawl hot topic and trending keyword data from social media platforms in real time. These social media platforms include Weibo, Twitter, Xiaohongshu, and Douyin. Specifically, open-source web crawler frameworks such as BeautifulSoup are used to build the crawler programs. By setting reasonable crawling strategies and request intervals, real-time hot keyword data such as "World Cup" and "earthquake" is obtained from social media platforms. At the same time, the crawler programs are deployed in a distributed architecture to improve the efficiency and stability of data crawling.

[0029] Furthermore, by calling the API interfaces of news media websites or news aggregation platforms, information related to news events can be obtained. It is understandable that each API interface usually provides rich news data, including news titles, content, publication time, etc. At the same time, the obtained news data is associated with the aforementioned hot word data to ensure the correlation between hot topics from different data sources and improve the efficiency of data processing.

[0030] In some embodiments, the association between hot word data and news events includes preliminary matching and event feature matching. Preliminary matching includes keyword matching and semantic association matching. Keyword matching is for hot word data in the format of a single keyword. Specifically, it involves matching the captured hot word data with the title and content of the news event. Specifically, string matching algorithms or regular expressions are used for matching. For example, if the hot word data is "Paris Games", then the hot word data is searched in the title and content of the news event. In addition, for hot word data in the format of short sentences, semantic association is used for preliminary matching. Specifically, natural language processing technology is used to calculate the semantic similarity between the hot word data and the news title of the news event. If the semantic similarity exceeds a preset threshold, the hot word data is considered to be related to the news event. In this process, the hot word data and the news title of the news event need to be converted into vectors, and then the cosine similarity between the two is calculated. The cosine similarity is used to measure the relevance between the hot word data and the news event. For event feature matching, specifically time and location feature association matching, the preliminarily matched hot word data and news events are encapsulated into a preliminary association data set. Each data entry in the preliminary association data set includes fields for hot word, news title, news content, original time information, and original location information. For time features, if both the hot word and the news event contain a clear and consistent time point, it is determined to be a strong time association. If the event referred to by the hot word is within the time range described by the news event, or vice versa, it is determined to be a time association. Further, time association inference is made based on the logical sequence between the content described by the hot word and the news event; if both the hot word and the news event... For activities with fixed cycles, temporal correlation is inferred based on their cyclical patterns. For location features, if both the hot word and the news event contain clear and consistent location names, a strong location correlation is determined. If the location in the hot word and the location in the news event have a hierarchical administrative relationship, a location correlation is determined. Furthermore, if the location of the news event falls within the area described by the hot word, a location correlation is determined. In addition, location correlation can be inferred based on the implicit geographical connections between entities or events in the hot word and the news event. If at least one of the temporal correlation and location correlation is determined to be correlated, the hot word and the news event are ultimately determined to have a spatiotemporal correlation.

[0031] In some embodiments, hot words and news events with spatiotemporal correlation are encapsulated into hot event data, which includes time, location, and hot topic, used to describe the time and location of the hot topic and the hot word data and news events associated with it. For hot words and news events without spatiotemporal correlation, they are encapsulated separately into hot event data, used to describe the time and location of the hot topic and the related hot word data or news events. Finally, a large amount of hot information obtained from social media platforms and news platforms is encapsulated into a hot event dataset, with hot topic ID as the key, to encapsulate and store the hot topics, which is beneficial for subsequent data extraction.

[0032] In some embodiments, the process of collecting user viewing behavior data on a television platform to obtain viewing preference characteristics includes: By acquiring users' viewing history, search records, and channel switching behavior on TV platforms, and using machine learning algorithms to segment user groups, construct user profiles, and obtain viewing preference characteristics of different user groups, we can achieve the desired results.

[0033] In this embodiment, the viewing history is obtained by recording the program ID, program type, program tag, viewing start time, viewing duration, and whether the program was played completely when the user requested or watched a replay. The search history specifically collects the user's search keywords, search time, and behavior after clicking on the search results on the TV platform. The channel switching behavior specifically records the live channel switching sequence, the duration of stay on each channel, and the switching time. In addition, the user's registration information, such as age, gender, and region, is also required to enhance the richness of the user profile.

[0034] Specifically, using the user's unique identifier as the key, the aforementioned viewing behavior data is associated and aggregated to form a user behavior data table. Each user corresponds to one record, and its fields are a statistical summary of all their behaviors. Furthermore, invalid / abnormal data and missing values ​​are processed in the aforementioned user behavior data table, such as deleting records with extremely short viewing durations or filling in certain missing static attributes.

[0035] Furthermore, the processed viewing behavior data undergoes feature processing. In some embodiments, the percentage of viewing time for different program types is calculated to obtain a viewing preference vector; the average daily viewing time, weekend / weekday viewing time ratio, and peak viewing periods are calculated to obtain viewing intensity features; high-frequency search terms are extracted using algorithms such as TF-IDF or TextRank to obtain search keyword features; the average channel switching frequency and preferred channel list are calculated to obtain channel switching features; and the user's viewing records are sorted by time to construct a program ID sequence or program type sequence to generate sequence features. Finally, all of the above features are concatenated into a high-dimensional numerical vector, representing a... The system analyzes users' overall behavioral characteristics and then uses machine learning algorithms, such as K-Means clustering and collaborative filtering, to segment users into different interest groups or market segments, such as sports enthusiasts, movie and TV show fans, and news and information followers. Furthermore, it labels each user with their respective group, such as sports enthusiasts, and outputs the preference strength of each user group for different program types, specific programs, and tags. For example, the preference weight for "live broadcasts" for the "sports enthusiasts" group is 0.9. In addition, for individuals within a group, collaborative filtering can be used to predict their interest scores for programs they haven't watched, thereby supplementing and refining their individual preference characteristics.

[0036] In this embodiment, the generated user profiles and viewing preference characteristics are beneficial for subsequently customizing exclusive channels or program schedules for different groups, thereby achieving more accurate program recommendations.

[0037] S200. Perform natural language processing on the hot event data to obtain hot keywords, calculate the similarity between the hot keywords and the metadata information of the program material database, and generate hot matching results. In this application, natural language processing is used on hot topic event data to reduce reliance on large amounts of labeled data, thereby reducing labeling costs, enhancing adaptability to new information, effectively handling a large amount of hot topic information updated daily, providing faster response speed, meeting real-time processing needs, while reducing reliance on rules manually written by domain experts, improving scalability and generalization ability, enabling keyword labeling of text to better adapt to different domains and contexts, and reducing reliance on domain expert knowledge.

[0038] In some embodiments, the step of performing natural language processing on the hot topic event data to obtain hot topic keywords includes: Based on preset scoring parameters, the hot topic subjects of the hot topic event data are segmented in multiple levels to obtain core words; By combining the weighted fusion formula, the comprehensive score of each core word is calculated; The core words are sorted according to the comprehensive score, and the top N core words are selected as hot keywords.

[0039] In this embodiment, the hot topic data is segmented into multiple levels, which can narrow the matching range of hot topic subjects containing a large number of words, reduce the interference of other non-keywords on hot topic keywords, and improve matching efficiency and accuracy. For the core words obtained after multi-level segmentation, a comprehensive score is calculated by combining a weighted fusion formula, and then the relevant hot topic keywords are obtained based on the comprehensive score. The hot topic keywords are used to obtain program materials related to the hot topic from the program material database and sort the TV programs.

[0040] In some embodiments, the weighted fusion formula is: Score(w) = m × TF-IDF(w) + (1-m) × Semantic_Score(w), where w represents the core word, TF-IDF represents the statistical feature, used to measure the importance of the core word in the current document and its scarcity in the entire corpus. Specifically, TF-IDF(w) = (number of times w appears in the current document / total number of words in the current document) × log2 (total number of documents in the corpus / number of documents containing w), and Semantic_Score(w) represents the semantic feature, used to measure the semantic importance of the core word. It is usually generated by a deep learning model, such as BERT, to capture the true meaning of the word in the current context. For example, in sports news, "apple" is more likely to refer to fruit, while in technology news it refers to a company. m is an adjustable parameter used to balance the weights of statistical features and semantic features. By adjusting the value of m, a more suitable matching strategy can be selected according to different application scenarios and needs.

[0041] In some embodiments, the hot topic subject includes hot word data and event text of news events, and the preset scoring parameters include paragraph scoring threshold, sentence scoring threshold and keyword scoring coefficient; The process involves multi-level segmentation of the hot topic subjects in the hot topic event data based on preset scoring parameters to obtain core keywords, including: The hot topics are cleaned, and the cleaned hot words are marked as the first core words. The cleaned event text is segmented into multiple paragraphs. The semantic vector model is used to analyze the semantic relevance of each paragraph to the event text, generate paragraph scores, and determine the core paragraphs by combining the paragraph score thresholds. Each core paragraph is segmented into multiple sentences. An importance scoring model is used to analyze the importance of each sentence in the corresponding paragraph, generate a sentence score, and determine the core sentences by combining the sentence score threshold. The core sentence is segmented into multiple words. A word score is generated based on the paragraph score of the paragraph in which the word is located, the sentence importance score of the sentence in which the word is located, and the comprehensive score of the word. The second core word is determined by combining the keyword score coefficient.

[0042] In this application, by combining factors such as word frequency and relational attributes, a multi-level segmentation method is adopted, which can more accurately extract keywords from long texts, meet the needs of text analysis and information retrieval in professional fields, improve the accuracy and precision of sentence segmentation strategies, optimize the segmentation process, avoid the problems of overly fine word segmentation and the inability to segment special words, and improve the accuracy and effectiveness of keyword extraction.

[0043] In this embodiment, the hot topic subject specifically includes hot word data and news events. Hot word data is usually in the format of a single keyword or a short sentence. Therefore, after text cleaning of the hot word data, such as removing irrelevant characters and word segmentation, relevant words can be obtained. These words can usually be used to represent the hot topic and are marked as the first core words. They are then combined with the second core words of subsequent news events to generate core words for subsequent hot topic keyword selection.

[0044] Furthermore, for event texts in news events, such as press releases and marketing articles, which are usually in long text data format, it is necessary to perform paragraph segmentation, sentence segmentation, and word segmentation in sequence to obtain individual words, score the words, and further determine the relevant core words.

[0045] In this embodiment, preset scoring parameters are introduced during the paragraph, sentence, and word segmentation processes. Specifically, these parameters include paragraph scoring thresholds, sentence scoring thresholds, and keyword scoring coefficients. These parameters are used to select core paragraphs, core sentences, and core words from the segmented paragraphs, sentences, or words. It is understood that because paragraphs contain a large amount of text content, the paragraph scores obtained through relevance calculations are relatively low. Conversely, the text content of sentences and words gradually decreases, and their scores obtained through relevance or importance calculations gradually increase. Therefore, the paragraph scoring thresholds, sentence scoring thresholds, and keyword scoring coefficients gradually increase, thus performing multi-level segmentation of the event text. For example, in one embodiment, the paragraph scoring threshold is 0.5, the sentence scoring threshold is 0.6, and the keyword scoring coefficient is 0.8. This achieves the gradual narrowing of event text containing a large amount of text content, thereby reducing the matching range, decreasing interference from other non-keyword options, and improving the matching efficiency and accuracy of hot keywords.

[0046] Specifically, the event text undergoes preprocessing operations such as text cleaning, including removing irrelevant advertisements, format tags, and special characters, and standardization. Then, the cleaned event text is segmented into multiple paragraphs according to natural paragraphs or semantic-based sentence segmentation models. Further, line break character or semantic vector models are used to analyze the semantic relevance of each paragraph to the event text, and a paragraph score is calculated for each paragraph. Paragraphs with scores higher than a threshold are identified as core paragraphs. For each core paragraph, sentence segmentation is performed, using punctuation such as periods and question marks or sentence recognition models to obtain multiple sentences. For the sentences within each core paragraph, [further details are needed]. An importance scoring model is used to calculate the importance of a sentence to its paragraph, generating a sentence score. Sentences with scores higher than the sentence score threshold are identified as core sentences. From these core paragraphs, core argument sentences, conclusion sentences, or factual sentences are extracted, further filtering out modifying and transitional sentences. Furthermore, for each core sentence, word segmentation techniques, such as jieba or BERT, are used to cut it into independent words or phrases, and a word score is generated for each word. Words with scores higher than the keyword score coefficient are identified as secondary core words. These are combined with the primary core words to generate core vocabulary for subsequent selection of hot keywords.

[0047] In some embodiments, during paragraph segmentation, natural paragraph separators in the text, such as line breaks and paragraph marks, are identified. By using tags or specific indentation formats, text is divided into multiple physical paragraphs. For continuous text lacking obvious format markers, a pre-trained semantic vector model is used for sentence-level analysis. Based on the semantic coherence of the context and the turning points of the topic, the text is divided into logically complete paragraphs. This ensures that a well-structured set of paragraphs can be obtained under any text format, laying a reliable data foundation for the subsequent selection of core paragraphs. For the semantic vector model, a deep learning semantic model, such as Sentence-BERT, SimCSE, or a finely tuned BERT model, is used to convert the title of the entire event text and each segmented paragraph into high-dimensional dense vectors. This semantic vector model can map semantically similar texts to similar positions in the vector space. Specifically, for each paragraph vector, the cosine similarity between it and the article title vector is calculated. The formula is: semantic similarity = (vector A · vector B) / (||vector A|| * ||vector B||). It can be understood that the closer the semantic similarity is to 1, the more relevant the paragraph is to the topic of the event text. Finally, the calculated semantic similarity is normalized and mapped to a preset score range as the paragraph score.

[0048] In some embodiments, for each core paragraph, a deep learning-based sentence recognition model is used for sentence segmentation. This model not only recognizes explicit punctuation such as periods, question marks, and exclamation marks, but also identifies abbreviations and headings. For example, it avoids misinterpreting points in "Dr.", "Fig.1," and "USA" as sentence ends. It can also correctly segment multiple sentences within quotation marks into multiple complete sentences within quotations or dialogues, and identify each segmented sentence as an independent semantic unit. This ensures the semantic integrity and independence of each sentence, facilitating accurate assessment of the importance of each sentence within its corresponding paragraph. For sentence scoring, an importance scoring model is used to calculate a score for each sentence to measure its importance. This sentence score is a multi-factor weighted fusion result, including positional, content, and semantic dimensions. For the positional dimension, the first and last sentences of a paragraph are assigned higher base weights, while a decay or enhancement function is defined to dynamically calculate weights based on the sentence's relative position within the paragraph. For the content dimension, named entities in the sentence are identified, and their density and importance are calculated. The importance of the content is measured by the density of named entities. Specifically, named entities include names of people, organizations, places, and times. , specifically, corresponding weights are assigned to each type of named entity, and then the sum of the entity type weights of the named entities in the sentence is calculated. From this, the entity naming density of the sentence can be calculated. It can be understood that by introducing entity type weights, the importance of each named entity is determined, and the entity type weights can be calculated and determined according to experience or the frequency of the entity appearing in the event text or the TF-IDF value; for the semantic dimension, a semantic vector model is used to calculate the cosine similarity between the sentence vector and the vector of the title of the paragraph to which it belongs. If there is no title, the vector of the core sentence of the paragraph is used instead. The higher the similarity, the more relevant the sentence is to the paragraph theme and the higher the score. The final sentence score = a × position dimension score + b × + c × semantic dimension score, where a, b, and c are configurable weight coefficients, and a + b + c = 1. Thus, the importance of the sentence can be evaluated more comprehensively and robustly, and the argument sentences, conclusion sentences, and factual sentences can be identified.

[0049] In some embodiments, a word segmentation tool, such as Jieba or LTP, is used for basic vocabulary segmentation. In addition, a domain-specific dictionary, such as a sports term dictionary or a scientific and technological proper noun dictionary, is loaded into the word segmentation tool to ensure that professional terms and emerging hot words can be correctly recognized as a complete word element, avoiding segmentation errors. At the same time, based on statistical rules or sequence annotation models, adjacent words are identified and merged to form more representative phrases. For example, "artificial" and "intelligence" are merged into "artificial intelligence", and "Beijing" and "sports meeting" are merged into "Beijing sports meeting", providing a high-quality and semantically complete candidate word set for subsequent core word selection. Further, the above candidate word set is filtered by word type, and only real-word candidate words such as nouns, verbs, gerunds, and professional terms are retained. For example, words with clear meanings such as "competition", "celebrate", and "victory" are retained, and stop words and function words such as "of", "in", and "so" are filtered out. At the same time, single-character words and overly long meaningless strings are removed, greatly reducing the number of invalid candidate words and the complexity of subsequent calculations. For word scoring, the specific calculation formula is S(w) = Score(w) × S_sentence(s) × S_paragraph(p), where S_sentence(s) represents the sentence score of the sentence s where the candidate word w is located, S_paragraph(p) represents the paragraph score of the paragraph p where the candidate word w is located, then S(w) represents the word score of the candidate word w, and Score(w) represents the importance of the candidate word w itself. The comprehensive score obtained through the above weighted fusion formula Score(w) = m × TF-IDF(w) + (1 - m) × Semantic_Score(w) is used to measure, where TF-IDF represents statistical features, Semantic_Score(w) represents semantic features, and m is an adjustable parameter used to balance the weights of statistical features and semantic features.

[0050] This solution, which combines sentence scoring, paragraph scoring, and word scoring, has significantly reduced the number of invalid candidate words and improved the efficiency of keyword acquisition.

[0051] Preferably, to more accurately obtain relevant program lists based on user preferences, the word score can also be obtained in the following way: S(w) = Score(w) × S_sentence(s) × S_paragraph(p) × S_interest(w), where S_interest(w) represents the similarity between the inherent attributes of candidate word w and the user's viewing preference features. The inherent attributes of candidate word w can be obtained based on authoritative data sources such as dictionaries, knowledge graphs, and encyclopedias. Alternatively, they can be obtained by calculating the cosine similarity between candidate word w and a series of known sports core words (such as "football" and "basketball") or entertainment core words (such as "movies" and "music"). Alternatively, a large number of texts with labeled preference features (such as sports texts, entertainment texts, and news texts) can be used to train a machine learning model (such as a Bayesian classifier, support vector machine SVM, or deep learning model BERT) to learn the features of various types of texts. Then, the candidate word w is input into the model, and the model outputs the probability of which type of text w belongs to. This probability can be used as S_interest(w).

[0052] This solution builds upon the previous one by incorporating user preference characteristics, making the subsequent acquisition of trending keywords and the generation of program order based on the trending keyword matching results more closely aligned with user preferences, thus improving the user experience.

[0053] In some embodiments, the program material database includes multiple program materials, each of which is provided with metadata information, including program type, theme content, broadcast duration, production time and copyright information; In this embodiment, a program material database is constructed to store and manage various types of program content. Specifically, the program material database contains multiple program materials, and detailed metadata information is set for each program material, including program type, theme, broadcast duration, production time, copyright information, etc. The program types include news, sports, entertainment, etc. Finally, the above-mentioned program material database is stored using a relational database (such as MySQL, PostgreSQL) or a non-relational database (such as MongoDB), and an efficient data indexing mechanism is established to enable quick querying and retrieval of program materials, thereby providing rich program material resources for program scheduling and ensuring that content matching hot events and user preferences can be selected from a large number of programs during the program scheduling process.

[0054] The step of calculating the similarity between the hot keywords and the metadata information of the program material database to generate hotspot matching results includes: The semantic similarity between the hot keywords and the program type and theme content of the metadata information is calculated to obtain preliminary program materials; Determine the emotional tendency of trending events, and adjust the preliminary program materials based on the emotional tendency to obtain related program materials; The BERT model is used to perform entity recognition on the hotspot event data to obtain hotspot entities; The associated program materials are linked with the hotspot entities to construct an event-program association graph. In the event-program association graph, one hotspot entity corresponds to one or more program materials, generating hotspot matching results.

[0055] Specifically, by performing precise semantic matching between real-time trending events and program material databases, the relevance of content is improved, ensuring that program schedules can reflect current hot topics in a timely manner, attracting audience attention and improving the timeliness and relevance of program content.

[0056] In this embodiment, semantic similarity is calculated between each hot keyword and the program type and theme content of the metadata information in the program material library. Specifically, each hot keyword is concatenated into a query text Q and converted into a query vector v_Q. For each program information in the program material library, a comprehensive metadata vector v_i is generated by combining its metadata information with the program type and theme content. The similarity between the query vector and each metadata vector is calculated as sim(Q,i)=cos(v_Q,v_i), and they are sorted according to the similarity. The Top-K programs are retained. Thus, through the above similarity calculation, preliminary program materials can be screened from the program material database.

[0057] Furthermore, the initial program material is adjusted based on the sentiment of trending events. For example, a negative trending topic like "disaster relief" should be matched with serious news reports or documentaries, rather than entertainment programs. Specifically, a pre-trained language model based on the Transformer architecture is used for sentiment analysis. This model is fine-tuned for domain adaptation on a large-scale sentiment-annotated corpus to optimize its ability to judge the sentiment of news and trending event texts. That is, product reviews, social media texts, etc. are used as training data to perform large-scale sentiment annotation, so that the model's final output is a sentiment polarity probability distribution, which usually contains three categories: positive, negative, and neutral. Each category corresponds to a confidence probability value, which ultimately determines the sentiment of the trending event.

[0058] Furthermore, entity recognition is performed on the hot topic event data, that is, entity recognition is performed on the hot topic subjects in the hot topic event data to obtain hot topic entities, which are used for the construction of subsequent event-program association graphs, thereby ensuring the accuracy of matching. Specifically, the hot topic event data is merged into hot topic event text according to time, location, hot words and events, and used as input to the BERT model. During the training process of the BERT model, entity labels are labeled for each hot topic event text, and the model is trained using the above labeled data to optimize the cross-entropy loss, so that the BERT model can predict entity labels and map the predicted entity labels back to the hot topic event text to extract the corresponding entities.

[0059] Specifically, a hot topic entity in an event-program association graph can typically be associated with multiple program materials. For example, a hot topic entity "Paris Games" can be associated with live broadcast materials, retrospective materials, documentary materials, and special feature materials. Specifically, live broadcast materials refer to the current live broadcast of the games, retrospective materials are compilations of classic moments from past games, documentary materials are documentaries introducing the history of Paris or the games, and special feature materials include news interviews with relevant athletes. Thus, by using the event-program association graph, we can uncover deeper and more diverse connections between hot topics and programs, thereby recommending richer and more comprehensive content combinations and obtaining corresponding hot topic matching results.

[0060] S300. Based on the viewing preference features and the hotspot matching results, an initial population is generated, and a genetic algorithm is used to optimize the initial population to generate the program arrangement order; In this application, a multi-objective genetic algorithm is used to optimize the program order, taking into account user preferences and content diversity, avoiding homogenized recommendations, improving the viewing experience and satisfaction of the audience, reducing human intervention, improving the efficiency and accuracy of program scheduling, and adapting to the needs of multiple scenarios.

[0061] In some embodiments, step S300 above includes: An initial population is generated based on the viewing preference features and the hotspot matching results; In the initial population, multiple programs are sequentially encoded into gene sequences, each of which includes a program ID, a playback time period, and a playback priority. Using user interest satisfaction, hot topic relevance, and content diversity as optimization objectives, a fitness function is constructed. Through selection, crossover, and mutation operations, the initial population is iteratively evolved until a preset iteration condition is met, resulting in an optimized program arrangement order.

[0062] In this embodiment, the program scheduling problem is modeled as an optimization problem. User interest satisfaction, hot topic relevance, and content diversity are the optimization objectives. A genetic algorithm is used as the optimization algorithm to optimize the program scheduling. In this process, the corresponding viewing preference characteristics are obtained based on the user's group. Combined with the hot topic matching results obtained by the event-program association graph in step S200 above, multiple program arrangement orders are randomly generated to construct an initial population.

[0063] Specifically, in the genetic algorithm, a population containing multiple program ordering sequences is initialized. Each individual represents a possible program schedule, i.e., a candidate program schedule. Within this initial population, each candidate program schedule is sequentially encoded into a gene sequence using chromosome encoding. Each gene sequence contains a program ID, playback time slot, and playback priority. Thus, one gene sequence represents a program ordering sequence. Furthermore, by constructing a fitness function, the initial population undergoes iterative evolution through genetic operations such as selection, crossover, and mutation. The selection operation involves retaining the top 20% of elite individuals in each iteration and using a roulette wheel to select the remaining 80%. The crossover operation mimics the crossing over of biological chromosomes, randomly selecting... Two program schedules are selected, and a portion of their time slots are swapped. For example, if program schedule A and program B have different program sequences at 8 PM, swapping the programs in this time slot can generate a descendant program schedule that retains the advantages of both A and B. Alternatively, the order of the programs in the program schedule can be swapped to generate a descendant program schedule. Mutation operations replace low-fitness programs with a 10% probability. Finally, the population is iterated and evolved until a predetermined number of iterations or convergence conditions are reached to obtain the final optimized population. The individual with the highest fitness in the population is selected as the optimized program schedule, and the program arrangement order is determined. It is understandable that the fitness function is used to exclude some abnormal individuals during the calculation and operation process, and to retain as many common individuals as possible to balance the objective.

[0064] Specifically, during the generation of program schedules for the initial population, relevant program materials, such as live broadcasts and news, are selected from the hotspot matching results and inserted into appropriate time slots in the program schedule. Based on viewing preference characteristics, programs are selected from the user's preferred program types to fill other time slots. The user's prime viewing time is determined based on their viewing intensity characteristics. Simultaneously, to ensure program schedule diversity, different individuals may vary in their program selections. In one embodiment, the user's viewing preference characteristics indicate a preference for sports, some interest in news, and a prime viewing time of 20:00-21:00. The current hotspot matching result is "Paris Games." Therefore, during the initial population generation process, sports-related programs, especially those related to the Paris Games, should be prioritized. For example, a program schedule might be arranged as follows: "19:00 Variety Show → 19:30 Classic Movies → 20:00 Late Night News → 20:30 Basketball Park → 21:00 Popular Drama → 21:30..." The "news" segment, as is understandable, presents issues in the initial population's program schedule, such as the placement of popular events during off-peak hours, scattered news programs, and insufficient sports content favored by users. Therefore, optimization is necessary. This is achieved by calculating the fitness score of the program schedule using a fitness function and then iterating it with other program schedules in the initial population through a "selection-crossover-mutation" process. For example, swapping the 20:00 football night from another program schedule with the 20:00 late-night news from the aforementioned program schedule generates a new descendant program schedule. Alternatively, programs in the program schedule can be swapped to concentrate news programs during prime time, generating a new descendant program schedule. The fitness score of each descendant program schedule is then calculated, and through multiple iterations, the individual with the highest fitness score in the population is ultimately selected as the optimized program schedule.

[0065] Furthermore, the fitness function comprehensively considers user interest satisfaction, hot topic relevance, and content diversity to evaluate the quality of each individual in the population, thereby optimizing the population. Specifically, the fitness function is: Fitness = α × (user interest matching degree) + β × (hot topic relevance) + γ × (content diversity entropy), where user interest matching degree is obtained through historical data prediction. Specifically, based on user profiles, the overall viewing interest probability of the program list is predicted. This can be achieved by first obtaining user interest tags, such as "sports enthusiast" or "movie and TV drama fan," and then obtaining the metadata of the program list, such as each program type and topic tags, based on the preference weights of these interest tags for different program types (such as sports, news, and variety shows) or specific program tags (e.g., the preference weight for "live broadcast" is 0.9). For example, a program might be {type: "sports", tags: ["football", "Champions League", "live broadcast"]}, and the user's preference weight for that program type can be directly retrieved. Calculate the similarity between the program tag set and the user preference tags, or directly use the pre-calculated tag weights, and weight and fuse the matching scores of type and tag to obtain the single interest score of the program. Then, aggregate the interest scores of all programs in the program list using a weighted average to obtain the overall viewing interest probability. Content diversity entropy is calculated through program type distribution. Specifically, the probability distribution of all program types in the program list is statistically analyzed, and then the Shannon entropy of this probability distribution is calculated. Hotspot relevance is the degree of correlation between programs in the program list and hot topics, which can be obtained by calculating the similarity between the thematic content of each program in the program list and hot keywords. α, β, and γ are weighting coefficients, and strategies can be flexibly adjusted according to the program list's positioning. Specifically, if the program list is positioned towards news and information, it needs to respond quickly to hot topics and act as an information source for the public, with lower demand for personalization; therefore, hotspot relevance has the highest weight, followed by user interest matching, and content diversity entropy has the lowest weight. If the program list is positioned towards general entertainment, it needs to balance user experience and content diversity, moderately paying attention to hot topics while satisfying popular tastes; therefore, user matching has the highest weight, followed by content diversity entropy, and hotspot relevance has the lowest weight. The user profile is obtained based on the user information collected in step S100.

[0066] S400: Monitors real-time changes in trending events and user preferences, and dynamically adjusts the program order accordingly.

[0067] In this embodiment, the program arrangement order obtained in step S300 is combined with the metadata information of the program materials in the program material database to generate a final program list. The final program list is then arranged and displayed according to chronological order and channel layout, generating a visual program list interface. Simultaneously, changes in trending events and user preferences are monitored in real time. When significant changes in trending events or substantial shifts in user preferences are detected, an automatic program list adjustment mechanism is triggered, re-invoking the genetic algorithm to adjust and update the program list. This provides the television platform with a real-time updated program list, ensuring that program content can respond promptly to changes in trending events and user needs, thereby improving program appeal and the viewer's viewing experience.

[0068] Please see Figure 2 As shown, the present invention also provides a program dynamic scheduling system, the system comprising: Data acquisition module 201: used to collect hot event data from social media platforms and news platforms in real time, and at the same time collect user viewing behavior data on TV platforms to obtain viewing preference characteristics; Semantic analysis module 202: used to perform natural language processing on the hot event data to obtain hot keywords, calculate the similarity between the hot keywords and the metadata information of the program material database, and generate hot matching results; Genetic algorithm optimization module 203: used to generate an initial population based on the viewing preference features and the hotspot matching results, and to optimize the initial population using a genetic algorithm to generate the program arrangement order; Dynamic adjustment module 204: Used to monitor changes in hot events and user preferences in real time, and dynamically adjust the program arrangement order.

[0069] It is understandable that, such as Figure 1 The content of the program dynamic scheduling method embodiments shown is applicable to the program dynamic scheduling system embodiments. The specific functions implemented by the program dynamic scheduling system embodiments are as follows: Figure 1 The program dynamic scheduling method shown in the embodiment is the same, and the beneficial effects achieved are the same as those described above. Figure 1 The beneficial effects achieved by the illustrated program dynamic arrangement method embodiment are also the same.

[0070] It should be noted that the information interaction and execution process between the above systems are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0071] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0072] Please see Figure 3 As shown, this embodiment of the invention also provides a computer device 3, including: a memory 302 and a processor 301, and a computer program 303 stored in the memory 302. When the computer program 303 is executed on the processor 301, it implements the program dynamic scheduling method as described in any of the above methods.

[0073] The computer device 3 may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that... Figure 3 The computer device 3 is merely an example and does not constitute a limitation on the computer device 3. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0074] The processor 301 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0075] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may be an external storage device of the computer device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 3. Furthermore, the memory 302 may include both internal and external storage units of the computer device 3. The memory 302 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory 302 can also be used to temporarily store data that has been output or will be output.

[0076] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the program dynamic scheduling method as described in any of the above methods.

[0077] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographic device / computer device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0078] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for dynamic program scheduling, characterized in that, include: Real-time data collection of trending events from social media and news platforms, as well as user viewing behavior data on TV platforms, to obtain viewing preference characteristics; Natural language processing is performed on the hot topic event data to obtain hot topic keywords. The similarity between the hot topic keywords and the metadata information of the program material database is calculated to generate hot topic matching results. Based on the viewing preference features and the hotspot matching results, an initial population is generated, and a genetic algorithm is used to optimize the initial population to generate the program arrangement order; The program schedule is dynamically adjusted based on real-time monitoring of trending events and changes in user preferences.

2. The method as described in claim 1, characterized in that, The real-time collection of trending event data from social media and news platforms includes: Using a distributed web crawler, hot word data is captured in real time from social media platforms, while news media API interfaces are called to obtain news events; The hot word data and the news events are associated and data is extracted to generate structured hot event data, which includes time, location and hot topic.

3. The method as described in claim 1, characterized in that, The process of collecting user viewing behavior data on the television platform to obtain viewing preference characteristics includes: By acquiring users' viewing history, search records, and channel switching behavior on TV platforms, and using machine learning algorithms to segment user groups, construct user profiles, and obtain viewing preference characteristics of different user groups, we can achieve the desired results.

4. The method as described in claim 2, characterized in that, The step of performing natural language processing on the hot topic event data to obtain hot topic keywords includes: Based on preset scoring parameters, the hot topic subjects of the hot topic event data are segmented in multiple levels to obtain core words; By combining the weighted fusion formula, the comprehensive score of each core word is calculated; The core words are sorted according to the comprehensive score, and the top N core words are selected as hot keywords.

5. The method as described in claim 4, characterized in that, The hot topic includes hot word data and event text of news events, and the preset scoring parameters include paragraph scoring threshold, sentence scoring threshold and keyword scoring coefficient; The process involves multi-level segmentation of the hot topic subjects in the hot topic event data based on preset scoring parameters to obtain core keywords, including: The hot topics are cleaned, and the cleaned hot words are marked as the first core words. The cleaned event text is segmented into multiple paragraphs. The semantic vector model is used to analyze the semantic relevance of each paragraph to the event text, generate paragraph scores, and determine the core paragraphs by combining the paragraph score thresholds. Each core paragraph is segmented into multiple sentences. An importance scoring model is used to analyze the importance of each sentence in the corresponding paragraph, generate a sentence score, and determine the core sentences by combining the sentence score threshold. The core sentence is segmented into multiple words. A word score is generated based on the paragraph score of the paragraph in which the word is located, the sentence importance score of the sentence in which the word is located, and the comprehensive score of the word. The second core word is determined by combining the keyword score coefficient.

6. The method as described in claim 1, characterized in that, The program material database includes multiple program materials, and each program material is equipped with metadata information, which includes program type, theme content, broadcast duration, production time and copyright information. The step of calculating the similarity between the hot keywords and the metadata information in the program material database to generate hotspot matching results includes: The semantic similarity between the hot keywords and the program type and theme content of the metadata information is calculated to obtain preliminary program materials; Determine the emotional tendency of trending events, and adjust the preliminary program materials based on the emotional tendency to obtain related program materials; The BERT model is used to perform entity recognition on the hotspot event data to obtain hotspot entities; The associated program materials are linked with the hotspot entities to construct an event-program association graph. In the event-program association graph, one hotspot entity corresponds to one or more program materials, generating hotspot matching results.

7. The method as described in claim 1, characterized in that, Based on the viewing preference features and the hotspot matching results, an initial population is generated, and a genetic algorithm is used to optimize the initial population to generate the program arrangement order, including: An initial population is generated based on the viewing preference features and the hotspot matching results; In the initial population, multiple programs are sequentially encoded into gene sequences, each of which includes a program ID, a playback time period, and a playback priority. Using user interest satisfaction, hot topic relevance, and content diversity as optimization objectives, a fitness function is constructed. Through selection, crossover, and mutation operations, the initial population is iteratively evolved until a preset iteration condition is met, resulting in an optimized program arrangement order.

8. A dynamic program scheduling system, characterized in that, include: Data acquisition module: used to collect real-time data on trending events from social media and news platforms, and at the same time collect user viewing behavior data on TV platforms to obtain viewing preference characteristics; Semantic analysis module: used to perform natural language processing on the hot event data to obtain hot keywords, calculate the similarity between the hot keywords and the metadata information of the program material database, and generate hot matching results; Genetic algorithm optimization module: used to generate an initial population based on the viewing preference features and the hotspot matching results, and to optimize the initial population using a genetic algorithm to generate the program arrangement order; Dynamic adjustment module: Used to monitor changes in trending events and user preferences in real time, and dynamically adjust the program arrangement order.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images