Video hotspot segment determination method, apparatus, and electronic device

Abstract

The application discloses a video hot spot segment determination method and device and electronic equipment. The method comprises the following steps: dividing multi-dimensional video data to obtain a plurality of candidate segment units; clustering processing of barrage data in the candidate segment unit is performed to obtain a plurality of barrage data groups; first index data is determined based on the barrage data group, and a hot spot score is determined according to the first index data and second index data, wherein the first index data is used to quantify the semantic consensus strength and the hot spot burst trend of the barrage, and the second index data is used to quantify the hot spot contribution degree of the program context and the hot spot intensity of the user interaction behavior; and the hot spot segment is determined from the plurality of candidate segment units according to the hot spot score. The application solves the technical problems of hot spot recognition lag, fragmented expression of barrage and difficulty in distinguishing differentiated hotness characteristics of program stages due to the fact that the competition type long video live program in the related art usually relies on manual spot broadcasting or post-event statistics to find hot spots.

Description

Technical Field

[0001] This application relates to the field of Internet live video processing, and more specifically, to a method, apparatus, and electronic device for determining video hotspot segments. Background Technology

[0002] With the development of internet live streaming technology and long-form video platforms, an increasing number of competitive programs are being broadcast live or in a near-live format. These programs typically include multiple stages such as contestant introductions, stage performances, judges' comments, voting, advancement, and elimination. Viewers provide real-time feedback during the live stream through features like bullet comments, likes, votes, and sharing. For program operations, promptly identifying trending segments based on viewer feedback and transforming them into short video clips, hashtags, cover text, and promotional headlines is a crucial aspect of program placement and dissemination.

[0003] However, hotspot segment identification in related technologies often suffers from the following drawbacks:

[0004] Firstly, the discovery of traditional trending segments typically relies on operators monitoring the broadcast for extended periods or conducting offline analysis based on metrics such as view counts and total comments after the live stream ends. However, due to the longer duration of competitive programs, and the fact that trending segments (such as contestants' highlight moments or judges' insightful comments) often generate a surge of comments within just a few seconds or tens of seconds, manual identification can easily miss the optimal timing for capturing these segments. This results in promotional materials not being generated in time within the trending segment processing window, potentially missing the golden period for dissemination.

[0005] Secondly, in terms of bullet screen processing, relevant technologies mostly employ keyword matching, sensitive word filtering, or simple sentiment analysis to perform coarse-grained statistics on bullet screens. However, bullet screens in competitive live streams are characterized by colloquial language, diverse expressions, and fragmented short sentences. Simple statistical methods struggle to group bullet screens with similar meanings but different expressions into the same category of trending topics, leading to distorted or missed trending topic identification. Furthermore, the density of bullet screens varies significantly across different program stages, making it difficult for simple statistical methods to adapt to this dynamic change.

[0006] Furthermore, different stages of competitive programs (such as performance climaxes, judges' comments, and voting) have different communication logics. Related technologies typically use the total number of bullet comments or interactions as a uniform criterion, ignoring the differences in attention given to different program stages and specific program entities (such as contestants or judges). For example, the voting stage naturally involves high-frequency interactions; indiscriminately classifying it as a hot topic could easily lead to misjudgment. Conversely, the emotional outbursts during performance climaxes, if not weighted, may be overwhelmed by low-weighted, regular interactions.

[0007] There is currently no effective solution to the above problems. Summary of the Invention

[0008] This application provides a method, apparatus, and electronic device for determining video hotspot segments, in order to at least solve the technical problems of delayed hotspot identification, fragmented bullet screen comments, and difficulty in distinguishing the different popularity characteristics of different program stages in long video live broadcasts of competition-type programs that usually rely on manual monitoring or post-event statistics to discover hotspots.

[0009] According to one aspect of the embodiments of this application, a method for determining video hotspot segments is provided, comprising: dividing multidimensional video data to obtain multiple candidate segment units; clustering the bullet screen data within the candidate segment units to obtain multiple bullet screen data groups; determining first indicator data based on the bullet screen data groups, and determining a hotspot score based on the first indicator data and second indicator data, wherein the first indicator data is used to quantify the semantic consensus strength and hotspot outbreak trend of bullet screens, the second indicator data is used to quantify the hotspot contribution degree of the program context and the hotspot intensity of user interaction behavior, and the hotspot score is used to quantify the potential popularity of the candidate segment units; and determining hotspot segments from the multiple candidate segment units based on the hotspot scores.

[0010] Optionally, the multidimensional video data is divided to obtain multiple candidate segment units, including: acquiring multidimensional video data, wherein the multidimensional video data includes at least live video data, video timeline information, program flow information, bullet screen data, and user interaction data; based on the video timeline information, the live video data is divided into multiple candidate segment units according to a preset time window, and each candidate segment unit is associated with corresponding program flow information, bullet screen data, and user interaction data.

[0011] Optionally, the bullet screen data within the candidate segment unit is clustered to obtain a bullet screen data group, including: preprocessing the bullet screen data within the candidate segment unit into text and converting the preprocessed bullet screen data into bullet screen semantic vectors; determining the clustering radius based on the bullet screen density, bullet screen growth rate, and the number of bullet screen data groups in the historical clustering results within the candidate segment unit; and clustering the bullet screen semantic vectors based on the clustering radius to generate a bullet screen data group containing a lifecycle parameter and a decay weight.

[0012] Optionally, the method further includes: determining the number of times a bullet screen data group appears within a continuous time window; extending the lifecycle parameter and increasing the decay weight based on the number of appearances, the number of users covered, and the growth rate of bullet screen metrics when the number of appearances is greater than or equal to a preset number; and reducing the decay weight based on a preset decay rule when the number of appearances is less than a preset number or when the bullet screen data group does not match the candidate segment unit.

[0013] Optionally, the first indicator data includes data group concentration and hotspot outbreak prediction value. The first indicator data is determined based on the barrage data group, including: determining the data group concentration based on the number of barrages in the largest barrage data group within the candidate segment unit and the total number of barrages within the candidate segment unit, wherein the data group concentration is used to reflect the semantic consistency of the barrage data within the candidate segment unit; and using a hotspot prediction model to analyze the data group concentration and temporal feature sequence to obtain the hotspot outbreak prediction value, wherein the hotspot outbreak prediction value is used to represent the probability that the candidate segment unit will enter a hotspot state within a future time window.

[0014] Optionally, the second indicator data includes program entity attention, program stage weight, bullet screen density growth value, user emotion intensity value, and user interaction growth value. The hot topic score is determined based on the first and second indicator data, including: matching candidate segment units with corresponding program entities and program stages based on program flow information and bullet screen data within the candidate segment unit to obtain a program entity association list and program stage identifier corresponding to the candidate segment unit; determining the program entity attention of the candidate segment unit based on the program entity association list, and determining the program stage weight of the candidate segment unit based on the program stage identifier; determining the bullet screen density growth value, user emotion intensity value, and user interaction growth value within the candidate segment unit; and determining the hot topic score of the candidate segment unit based on data group concentration, hot topic outbreak prediction value, program entity attention, program stage weight, bullet screen density growth value, user emotion intensity value, and user interaction growth value.

[0015] Optionally, hotspot segments are determined from multiple candidate segment units based on hotspot scores, including: when the hotspot score is greater than a preset hotspot threshold, the candidate segment unit is determined as a hotspot segment; when the hotspot scores of multiple consecutive candidate segment units are all greater than the preset hotspot threshold, the multiple candidate segment units are merged into one hotspot segment, and the start and end times of the hotspot segment are adjusted according to the peak positions of the hotspot scores of the multiple candidate segment units.

[0016] Optionally, the method further includes: extracting hot words and phrases from the bullet screen data group corresponding to the hot segments according to preset filtering conditions; generating a list of candidate hot words and phrases based on the hot words and phrases; and generating promotional materials for the competition-type live broadcast program corresponding to the multi-dimensional video data based on the program entity attention, program stage weight, and the list of candidate hot words and phrases.

[0017] According to another aspect of the embodiments of this application, a video hotspot segment determination device is also provided, comprising: a segmentation module for segmenting multidimensional video data to obtain multiple candidate segment units; a clustering module for clustering the bullet screen data within the candidate segment units to obtain multiple bullet screen data groups; a first determination module for determining first indicator data based on the bullet screen data groups, and determining a hotspot score based on the first indicator data and second indicator data, wherein the first indicator data is used to quantify the semantic consensus strength of bullet screens and the hotspot outbreak trend, the second indicator data is used to quantify the hotspot contribution degree of the program context and the hotspot intensity of user interaction behavior, and the hotspot score is used to quantify the potential popularity of the candidate segment units; and a second determination module for determining hotspot segments from the multiple candidate segment units based on the hotspot score.

[0018] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory and a processor, wherein the memory is used to store program instructions; and the processor is connected to the memory and used to execute the above-described video hotspot segment determination method.

[0019] According to another aspect of the embodiments of this application, a non-volatile storage medium is also provided, the non-volatile storage medium including a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned video hotspot segment determination method by running the computer program.

[0020] According to another aspect of the embodiments of this application, a computer program product is also provided, including computer instructions, which, when executed by a processor, implement the above-described method for determining video hotspot segments.

[0021] In this embodiment, multiple candidate segment units are obtained by dividing multidimensional video data; multiple bullet screen data groups are obtained by clustering the bullet screen data within the candidate segment units; a first indicator data is determined based on the bullet screen data groups, and a hotspot score is determined based on the first indicator data and the second indicator data. The first indicator data is used to quantify the semantic consensus strength and hotspot outbreak trend of bullet screens, the second indicator data is used to quantify the hotspot contribution degree of the program context and the hotspot intensity of user interaction behavior, and the hotspot score is used to quantify the potential popularity of the candidate segment units; hotspot segments are determined from multiple candidate segment units based on the hotspot score, achieving the goal of real-time and accurate positioning of high-popularity segments in live programs. This achieves the technical effect of improving the timeliness of hotspot discovery and optimizing the accuracy of hotspot judgment, thereby solving the technical problems of delayed hotspot identification, fragmented bullet screen expression, and difficulty in distinguishing the differentiated popularity characteristics of program stages in competition-type long video live programs in related technologies, which usually rely on manual monitoring or post-event statistics to discover hotspots. Attached Figure Description

[0022] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0023] Figure 1 This is a hardware structure diagram of a computer terminal for implementing a video hotspot segment determination method according to an embodiment of this application;

[0024] Figure 2 This is a flowchart of a video hotspot segment determination method according to an embodiment of this application;

[0025] Figure 3 This is a flowchart of another method for determining video hotspot segments according to an embodiment of this application;

[0026] Figure 4 This is a structural diagram of a video hotspot segment determination device according to an embodiment of this application. Detailed Implementation

[0027] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0028] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0029] First, some nouns or terms that appear in the explanation of the embodiments of this application shall be interpreted as follows:

[0030] Competition-style live streaming programs: These refer to live streaming content centered around competitions and selections, such as singing talent shows, dance competitions, and band competitions. These programs are characterized by strong competition, highly segmented formats (such as performances, voting, and commentary), and significant fluctuations in audience emotions.

[0031] Candidate segment unit: In this application, it refers to the basic time block into which a continuous live video stream is divided according to a preset time window (such as a fixed duration or a sliding step size). Each unit contains a specific start and end time, associated program stage information, and corresponding bullet comments and interaction data, and is the smallest granular object for subsequent hotspot calculation.

[0032] Program Entities: Refers to specific objects or event elements in competitive live programs that have independent identifiable significance, attract audience attention, and can trigger specific discussions, such as contestant names, group names, judges, specific songs, etc.

[0033] Cluster radius: In clustering algorithms, this refers to the distance threshold used to determine whether two data points are considered "neighbors". In this application, the radius is not fixed, but dynamically adjusted based on the current window's bullet screen density, growth rate, and the number of existing similar groups: the radius is reduced at high density to avoid erroneous merging, and expanded at low density to merge synonyms.

[0034] Promotional materials: These refer to content components generated based on trending clips for social media dissemination, including clip titles, hashtags, cover text, and recommended placement strategies (such as placement time, channels, and target audience).

[0035] To address the issue of delayed hotspot identification in related technologies, this application provides a method for determining video hotspot segments. This method can be implemented in... Figure 1 The computer terminal shown is described below.

[0036] The video hotspot segment determination method provided in this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal for implementing a method for determining video hotspot segments is shown. Figure 1As shown, the computer terminal 10 may include one or more processors (shown as 102a, 102b, ..., 102n in the figure) (the processor may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission module 106 for communication functions connected via wired and / or wireless networks. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, and a BUS bus. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0037] It should be noted that the aforementioned one or more processors and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10. As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0038] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the video hotspot segment determination method in this embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned video hotspot segment determination method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0039] The transmission module 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission module 106 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission module 106 may be a radio frequency (RF) module, used for wireless communication with the Internet.

[0040] The display can be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10.

[0041] It should be noted here that, in some optional embodiments, the above... Figure 1 The computer terminal shown may include hardware elements (including circuitry), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. It should be noted that... Figure 1 This is only one instance of a specific particular instance, and is intended to illustrate the types of components that may exist in the aforementioned computer terminal.

[0042] In the above operating environment, this application provides an embodiment of a video hotspot segment determination method. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0043] The video hotspot segment determination method provided in this application embodiment is applicable to long video live streaming platforms, program operation platforms, video content processing platforms, or recommendation and distribution systems, and is especially applicable to live streaming competition programs such as singing, dancing, rapping, band performances, and talent shows.

[0044] Figure 2 This is a flowchart of a video hotspot segment determination method according to an embodiment of this application, such as... Figure 2 As shown, the method includes the following steps:

[0045] Step S202: Divide the multidimensional video data to obtain multiple candidate segment units.

[0046] Step S204: Cluster the bullet screen data within the candidate segment unit to obtain multiple bullet screen data groups, wherein the bullet screen data groups are used to represent a set of bullet screens that are semantically similar and located in the same time period.

[0047] Step S206: Determine the first indicator data based on the bullet screen data group, and determine the hot spot score based on the first indicator data and the second indicator data. The first indicator data is used to quantify the semantic consensus strength of bullet screens and the hot spot outbreak trend, the second indicator data is used to quantify the hot spot contribution degree of the program context and the hot spot intensity of user interaction behavior, and the hot spot score is used to quantify the potential popularity of candidate segment units.

[0048] Step S208: Determine hotspot segments from multiple candidate segment units based on hotspot scores.

[0049] Through steps S202 to S208, the goal of locating high-popularity segments in live broadcasts in real time and accurately is achieved. This improves the timeliness of hot topic discovery and optimizes the accuracy of hot topic judgment, thereby solving the technical problems of delayed hot topic identification, fragmented bullet screen comments, and difficulty in distinguishing the differentiated popularity characteristics of different program stages in long-form competitive live broadcasts, which typically rely on manual monitoring or post-event statistics to discover hot topics. A detailed explanation follows.

[0050] In step S202 above, the aim is to divide the continuous live stream into discrete units by using a fixed or sliding time window, and associate time and context metadata to provide a standardized data foundation for subsequent fine-grained hotspot analysis.

[0051] Optionally, the multidimensional video data is divided to obtain multiple candidate segment units, including: acquiring multidimensional video data, wherein the multidimensional video data includes at least live video data, video timeline information, program flow information, bullet screen data, and user interaction data, and the video timeline information includes at least the start and end times of the live broadcast of the competition-type live program, the start and end times of the segment, the real-time playback time, and the video frame timestamp; based on the video timeline information, the live video data is divided into multiple candidate segment units according to a preset time window, and each candidate segment unit is associated with corresponding program flow information, bullet screen data, and user interaction data.

[0052] In this embodiment, the multidimensional video data includes at least one type of live competition program's live video data, video timeline information, program flow information, bullet screen data, and user interaction data. The live video data serves as the visual carrier of the live competition program; the video timeline information identifies the time progress, with fields including the start and end times of the live stream, real-time playback time, video frame timestamps, and the start and end times of the segments to be divided; the program flow information describes the program's content structure and is typically provided by a live streaming control console, program schedule, or manual annotation system, including program stage identifiers (such as performance climaxes, judge comments, voting segments, advancement / elimination, etc.), associated program entity identifiers (such as contestant names, group names, song identifiers, judge identifiers), and specific event identifiers; the bullet screen data reflects real-time audience feedback, including bullet screen text content, sending timestamps, sending user identifiers, and live stream identifiers; and the user interaction data quantifies audience participation in the live stream, including records of actions such as liking, commenting, voting, sharing, collecting, following, or replaying.

[0053] Furthermore, based on the acquired video timeline information, a preset time window strategy (supporting both fixed-duration and sliding-duration window modes) can be used to segment continuous live video data and generate multiple discrete candidate segment units.

[0054] Preferably, to balance the real-time performance of hotspot detection with the accuracy of boundary identification, a window length of 30 seconds and a sliding step of 5 seconds can be used for segmentation. For example, in fixed window mode, a non-overlapping segment is generated every 30 seconds; in sliding window mode, a new 30-second segment is generated every 5 seconds, resulting in a 25-second time overlap between adjacent segments. This improves the accuracy of identifying the start and end points of hotspot segments and avoids truncation of hotspot content due to improper window segmentation.

[0055] Finally, a data index is created for each generated candidate segment unit, and it is associated with the corresponding program flow information, bullet screen data, and user interaction data to form a structured candidate segment unit object. Through this association process, each candidate segment unit not only contains a video time segment, but also carries rich semantic context and interaction data, enabling subsequent hotspot analysis to make comprehensive judgments based on program flow and audience feedback.

[0056] In this step, by acquiring multi-dimensional video data and dividing candidate segment units according to a preset window and associating them with contextual information, the continuous live stream is discretized into data units with uniform temporal granularity and rich semantic background, providing a standardized data foundation for subsequent accurate and real-time hotspot identification.

[0057] In step S204 above, the aim is to normalize, vectorize, and dynamically cluster the bullet comments to generate similar bullet comment groups with lifecycle management (i.e., the bullet comment data groups mentioned above), so as to accurately capture the core consensus expressions of the audience in the same time period and filter out outdated interference.

[0058] In this embodiment, the bullet screen data within candidate segment units is clustered to obtain bullet screen data groups. This includes: preprocessing the bullet screen data within candidate segment units and converting the preprocessed bullet screen data into bullet screen semantic vectors; determining the clustering radius based on the bullet screen density, bullet screen growth rate, and the number of bullet screen data groups in historical clustering results within the candidate segment units, wherein the clustering radius is inversely proportional to the bullet screen density; and clustering the bullet screen semantic vectors based on the clustering radius to generate bullet screen data groups containing a lifecycle parameter and a decay weight. The lifecycle parameter represents the existence state of the bullet screen data group in the time series, and the decay weight is used to suppress the weights of bullet screen data groups that do not match the candidate segment units. The specific implementation process is as follows:

[0059] 1. Text preprocessing.

[0060] The raw bullet comments captured within candidate segment units undergo data cleaning and normalization to eliminate noise and standardize the expression format for subsequent semantic analysis. For example, invalid symbols (such as special characters and garbled text) are removed; duplicate characters are compressed; emoticons are converted into standard sentiment markers; synonym replacement and colloquial expression standardization are performed; and sensitive words and abnormal spam bullet comments are filtered out. The preprocessed bullet comment text is cleaner and semantically standardized, laying the foundation for generating high-quality semantic vectors.

[0061] 2. Semantic vectorization.

[0062] By using a pre-trained natural language processing model, the pre-processed bullet screen text is converted into bullet screen semantic vectors in a high-dimensional space. In this vector space, bullet screens with similar semantics are closer in geometric distance, enabling the computer to quantify the semantic similarity between different bullet screens and providing data support for subsequent clustering operations.

[0063] 3. Determine the dynamic cluster radius.

[0064] To ensure that the clustering results can adapt to the audience's expression characteristics under different barrage densities, the clustering radius of the clustering algorithm can be dynamically calculated and determined based on the barrage density, barrage growth rate, and the number of barrage data groups in the historical clustering results within the current candidate segment unit.

[0065] It should be noted that the clustering radius is inversely proportional to the current window's bullet screen density: when the bullet screen density is high or the growth rate is fast (such as during the climax of a performance), the system will automatically reduce the clustering radius and increase the similarity threshold to prevent the incorrect merging of bullet screens with different semantics but coincidental words, thus avoiding overly coarse clustering granularity; when the bullet screen density is low (such as during the prelude or lull period), the system will appropriately increase the clustering radius and lower the similarity threshold, merging bullet screens with similar semantics but slightly different expressions (such as "highlight", "too bright", "highlight moment") into the same group to prevent synonyms from being overly split and to ensure the adaptability of the clustering granularity.

[0066] 4. Clustering process to generate bullet screen data groups.

[0067] Based on a defined dynamic clustering radius, clustering operations are performed on the semantic vectors of bullet comments to generate multiple similar groups (i.e., bullet comment data groups). Each bullet comment data group not only contains the core phrase (the most representative text) within the group, but also maintains two key attributes: a lifecycle parameter and a decay weight. The lifecycle parameter is used to quantify the existence state of the bullet comment data group in the time series, which can distinguish between active expressions and transient noise; the decay weight is a dynamic adjustment coefficient used to suppress the weights of bullet comment data groups that do not match the current candidate segment unit.

[0068] In the above implementation process, by combining text preprocessing and semantic vectorization with dynamic clustering radius and life cycle decay mechanism, it is possible to accurately merge semantically similar bullet screen expressions under different bullet screen densities, and effectively suppress the interference of outdated or irrelevant expressions, thereby improving the ability of bullet screen similarity groups to respond to real-time concentrated feedback from the audience and the accuracy of hotspot calculation.

[0069] In this embodiment, a time-series-based dynamic weight adjustment mechanism is introduced to finely manage the lifecycle and influence of bullet screen data groups. This mechanism monitors the performance of bullet screen data groups within a continuous time window, dynamically updating their lifecycle parameters and decay weights to ensure that hotspot calculations focus on currently active and influential audience expressions. This includes: determining the number of times a bullet screen data group appears within a continuous time window; extending the lifecycle parameter and increasing the decay weight based on the number of appearances, the number of covered users, and the bullet screen metric growth rate, where the number of covered users represents the number of users who have sent bullet screens within the data group, and the bullet screen metric growth rate represents the rate of change of the bullet screen data group compared to the previous time window; and reducing the decay weight according to a preset decay rule if the number of appearances is less than the preset number or if the bullet screen data group does not match the candidate segment unit.

[0070] Specifically, the system counts the number of times each bullet comment data group appears within a continuous time window in real time. When the number of times a bullet comment data group appears is greater than or equal to a preset threshold (e.g., it appears in 3 consecutive windows, or the cumulative number of appearances exceeds 5 times), the expression is determined to have sustained popularity and importance, and the following operations are performed:

[0071] 1) Extend lifecycle parameter: Mark the lifecycle status of the bullet screen data group as "active" or "continuous" to extend its effective existence time, so that it can continue to be eligible for calculation in multiple subsequent windows and avoid being accidentally deleted due to short window fluctuations;

[0072] 2) Increase decay weight: Dynamically increase the decay weight coefficient based on the frequency of occurrence, the number of users covered, and the metric growth rate. The more frequent the occurrence, the more users covered, and the greater the metric growth rate, the greater the increase in weight.

[0073] When the number of occurrences of a bullet screen data group is less than the preset number (popularity decline), or when the bullet screen data group does not match the program stage identifier and associated program entity of the candidate segment unit, the attenuation weight is reduced according to the preset attenuation rules, such as exponential attenuation or linear reduction, to prevent old hot words from interfering with the current new hot topics and to ensure the timeliness and accuracy of hot topic judgment.

[0074] As the decay weight gradually approaches zero, the contribution of this bullet screen data group to the subsequent hot topic score calculation formula also decreases significantly, until it completely exits the hot topic calculation.

[0075] In the above implementation process, by dynamically adjusting the life cycle and decay weight of the bullet screen data group based on the frequency of occurrence, number of users covered, growth rate and context matching degree, the weight of continuously rising hot topics is strengthened and outdated or irrelevant topics are automatically suppressed. This effectively reduces the interference of historical bullet screens on the current hot topic judgment and improves the accuracy and timeliness of hot topic identification.

[0076] In step S206 above, the aim is to integrate intrinsic indicators (consensus and trends) based on the bullet screen data group and extrinsic indicators (user interaction and program context) based on the context to calculate an adaptive hotspot score, so as to accurately quantify the real-time potential popularity of candidate segments.

[0077] In this embodiment of the application, the first indicator data includes data group concentration and hotspot outbreak prediction value.

[0078] Optionally, the first indicator data is determined based on the bullet screen data group, including: determining the data group concentration based on the number of bullet screens in the largest bullet screen data group within the candidate segment unit and the total number of bullet screens within the candidate segment unit, wherein the largest bullet screen data group is used to represent the bullet screen data group containing the most bullet screens within the candidate segment unit, and the data group concentration is used to reflect the semantic consistency of the bullet screen data within the candidate segment unit; a hotspot prediction model is used to analyze the data group concentration and the temporal feature sequence to obtain a hotspot outbreak prediction value, wherein the temporal feature sequence includes at least one of the following: a sequence of bullet screen counts within multiple preset sliding windows, a sequence of user emotion intensity, a sequence of user interaction behavior, a sequence of program entities, and a sequence of program stage identifiers, and the hotspot outbreak prediction value is used to represent the probability that the candidate segment unit will enter a hotspot state within a future time window. The specific implementation process is as follows:

[0079] First, the data set concentration is calculated. This indicator reflects the concentration and semantic consistency of audience expression within the candidate segment unit. The calculation method is as follows:

[0080] Retrieve all bullet comment data groups within the current candidate segment unit, and select the group containing the most bullet comments, i.e., the largest bullet comment data group. Then, calculate the ratio of the number of bullet comments in this largest bullet comment data group to the total number of bullet comments in the current candidate segment unit to obtain the data group concentration. The specific expression is as follows:

[0081]

[0082] Where C represents the central tendency of the data set. The number of bullet comments in the largest bullet comment data group. This represents the total number of bullet comments within the candidate segment unit.

[0083] It should be noted that the concentration C of the data group ranges from [0,1]. When the C value is high, it indicates that a large number of viewers are concentrated in expressing similar opinions or emotions, with a high degree of semantic consistency, which usually corresponds to strong resonance or controversy; when the C value is low, it indicates that the bullet comments are scattered, with diverse opinions, and the overall popularity may be relatively flat.

[0084] Secondly, a hotspot outbreak prediction value is generated. This indicator data is used to predict the probability that candidate segment units will enter a hotspot state within a short time window in the future. Specifically, a pre-trained hotspot prediction model (such as a model based on LSTM, GRU, or Transformer temporal network) can be used for processing. The input features of this model include at least M preset sliding windows of data group concentration sequences and other temporal feature sequences, such as:

[0085] 1) The sequence of bullet comments is used to reflect the time-varying trend of bullet comment traffic;

[0086] 2) User emotion intensity sequence, used to reflect changes in the intensity of audience emotions;

[0087] 3) User interaction behavior sequence, used to reflect the temporal changes of interactive behaviors such as liking, voting, and sharing;

[0088] 4) Program entity sequence, used to reflect the changes in the popularity of specific contestants, judges or songs on the timeline;

[0089] 5) Program stage identifier sequence, used to reflect the temporal transition status of program process stages (such as from performance to voting).

[0090] The hotspot prediction model models and analyzes the concentration sequence and other time-series characteristic sequences of the aforementioned data sets to capture the slope, inflection points, and periodic patterns of data changes, ultimately outputting a hotspot outbreak prediction value P. This value represents the probability that the candidate segment unit will become a hotspot within the next K seconds (or the next window).

[0091] It should be noted that if the P-value exceeds the preset hotspot prediction threshold, the candidate segment unit can be marked as a "predicted hotspot segment" or "candidate hotspot segment" in advance, so as to start the material generation process before the hotspot actually breaks out and seize the opportunity for dissemination.

[0092] In the aforementioned real-time process, the semantic consistency of the current bullet comments is quantified by calculating the data set, and multi-dimensional temporal features are analyzed using a hotspot prediction model to generate hotspot outbreak prediction values. This achieves accurate measurement of the concentrated trend of audience expression and early prediction of future hotspot outbreaks, thereby improving the timeliness and foresight of hotspot discovery.

[0093] In this embodiment of the application, the second indicator data includes program entity attention, program stage weight, bullet screen density growth value, user emotion intensity value, and user interaction growth value.

[0094] Optionally, the hot topic score is determined based on the first and second indicator data, including: matching the candidate segment unit with the corresponding program entity and program stage based on the program flow information and bullet screen data within the candidate segment unit, obtaining a list of program entity associations and program stage identifiers corresponding to the candidate segment unit; determining the program entity attention of the candidate segment unit based on the program entity association list, and determining the program stage weight of the candidate segment unit based on the program stage identifier, wherein the program entity attention reflects the discussion heat of each program entity in the program entity association list, and the program stage weight reflects the contribution of different program stages to the formation of hot topics; determining the bullet screen density growth value, user emotion intensity value, and user interaction growth value within the candidate segment unit; and determining the hot topic score of the candidate segment unit based on data group concentration, hot topic outbreak prediction value, program entity attention, program stage weight, bullet screen density growth value, user emotion intensity value, and user interaction growth value. The specific implementation process is as follows:

[0095] 1. Perform program context matching to determine the corresponding program entity attention and program stage weight.

[0096] First, the program flow information associated with the candidate segment unit (including program stage identifiers, entity identifiers such as contestants / groups / songs) and the bullet screen data are read. By analyzing the entity words in the bullet screen data (such as contestant nicknames, team names, judge names), the candidate segment unit is matched with specific program entities to generate a list containing all mentioned entities within the segment, i.e., the program entity association list.

[0097] Secondly, obtain the program stage identifier (such as "Performance - High Note", "Judges' Comments", "Voting Session") of the current candidate segment unit from the program flow information.

[0098] Finally, based on the program entity association list, the frequency of mention of each program entity in the bullet comments, the total weight of related similar groups, or the number of users covered are statistically analyzed to derive the program entity attention score R, which reflects the audience's level of interest in a specific person or event. Additionally, based on the program stage identifier, the corresponding program stage weight S is obtained from a pre-set weight mapping table. Different stages contribute differently to the formation of trending topics; for example, the "performance climax" stage is usually assigned a higher base weight because this stage is more likely to produce segments with high dissemination value, while the "waiting for the result" stage may have a lower weight.

[0099] 2. Calculate the growth value of bullet screen density, user emotion intensity, and user interaction corresponding to user interaction.

[0100] First, calculate the ratio of the number of bullet comments in the current candidate segment unit to the average number of bullet comments in the previous N consecutive candidate segment units, which is the bullet comment density growth value D. This value reflects the degree of surge in the number of bullet comments. The larger the bullet comment density growth value, the more intense the bullet comment explosion.

[0101] Secondly, based on the average sentiment value of all bullet screen data groups within the candidate segment unit (which can be obtained through sentiment analysis models), the overall sentiment intensity is calculated using weighted averages to obtain the user sentiment intensity value E. High sentiment intensity usually corresponds to strong resonance or controversy, which is an important characteristic of trending topics.

[0102] Finally, the number or growth rate of non-bullet screen interaction behaviors (such as likes, votes, shares, collections, etc.) within the candidate segment unit are statistically analyzed and compared with the historical baseline. The user interaction growth value I is used to reflect the actual enthusiasm of the audience.

[0103] 3. Adaptive hotspot score calculation.

[0104] By combining the data from the first indicator (data group concentration C, hotspot outbreak prediction value P) and the second indicator (program entity attention R, program stage weight S, bullet screen density growth value D, user emotion intensity value E, user interaction growth value I), a weighted summation formula is used to calculate the hotspot score (H). The specific calculation formula is as follows:

[0105]

[0106] in, to These are adaptive weighting coefficients for each indicator, which can be dynamically adjusted based on the current program stage, data distribution characteristics, and historical promotion feedback. For example:

[0107] During the climax of a performance, the audience's emotional resonance and consistency in expression are key to identifying the hottest moments, and can automatically improve [their performance]. (Data group concentration) (Hotspot outbreak prediction value) and The weight of (user emotional intensity value) is given because emotional outbursts at this point are a better predictor of dissemination potential than simple interactive behavior.

[0108] During the voting phase, collective fan activity and the popularity of specific contestants dominate the formation of trending topics, which can automatically increase [the likelihood of success]. (User interaction growth value, especially votes and likes) and The weight of (the program's actual popularity) is because at this time, the hot topic is often closely related to the vote rate of a specific contestant or the ability to mobilize fans.

[0109] During the judges' comments phase, the clash of opinions and mentions of specific entities in the comments section are sources of discussion and could be appropriately increased. (Program's actual viewership, such as the number of times judges or contestants are mentioned) and The weight of (data group concentration, reflecting differences or consensus).

[0110] Furthermore, before being included in the calculation, each indicator can be transformed to a preset range using a piecewise function, sigmoid function, softmax function, or normalization mapping function to avoid bias in the hotspot score calculation results caused by different data volumes. Each weight coefficient can satisfy the following conditions:

[0111]

[0112] In the above implementation process, by combining intrinsic indicators such as semantic concentration and predicted value, as well as extrinsic context indicators such as program entities, stage weights and behavioral growth, and by adopting an adaptive weighting mechanism, multi-dimensional and dynamic hotspot score calculation was achieved. This enabled hotspot determination to accurately adapt to the dissemination patterns of different stages of competitive live broadcasts, significantly improving the accuracy and applicability of hotspot identification.

[0113] In step S208 above, the aim is to determine official hotspots or candidate hotspots based on hotspot scores, and to merge continuous segments to efficiently filter out specific hotspot segments with high dissemination potential from massive amounts of data.

[0114] In this embodiment, determining a hotspot segment from multiple candidate segment units based on hotspot scores includes: identifying a candidate segment unit as a hotspot segment when the hotspot score is greater than a preset hotspot threshold; merging multiple candidate segment units into one hotspot segment when the hotspot scores of multiple consecutive candidate segment units are all greater than the preset hotspot threshold, and adjusting the start and end times of the hotspot segment according to the peak positions of the hotspot scores of the multiple candidate segment units. The specific implementation process is as follows:

[0115] On the one hand, the determination of formal hotspot segments is carried out.

[0116] First, the hot topic score calculated for each candidate segment is monitored in real time and compared with a preset hot topic threshold. If the hot topic score of a candidate segment is greater than the preset hot topic threshold, the candidate segment is marked as an official hot topic segment, indicating that the segment has significant hot topic intensity or audience resonance at the current moment.

[0117] Secondly, hotspot segments are merged and their timing is corrected. Since the live stream is continuous and hotspots often last for several windows, the hotspot status of adjacent candidate segment units is monitored: when multiple consecutive candidate segment units are detected to have hotspot scores greater than a preset hotspot threshold, these segments are determined to belong to the same hotspot event, and these multiple consecutive candidate segment units are merged into a complete hotspot segment to avoid cutting the same hotspot into multiple fragmented small segments.

[0118] After merging, the start and end times of the merged hotspot segments are further adjusted based on the peak position of the hotspot scores in these consecutive segment units. For example, if the peak occurs in the second window, the system may extend forward by a certain proportion of time to capture the initial setup of the hotspot and extend backward to capture the lingering heat of the hotspot, thereby ensuring that the captured video segments fully contain the context before and after the hotspot outbreak, improving the coherence and attractiveness of subsequent promotional materials.

[0119] On the other hand, in order to intervene in the promotion of hot topics earlier, it is also possible to determine and confirm candidate hot topic segments (predicted hot topic segments).

[0120] When a candidate segment unit's hotspot score is greater than a preset candidate threshold (which is lower than the preset hotspot threshold), or its hotspot outbreak prediction value is higher than the preset outbreak threshold, the candidate segment unit is marked as a candidate hotspot segment (in a pending confirmation state). This usually occurs in the early stages of a hotspot outbreak, even though the current absolute score has not yet reached the official standard, but trend indicators suggest that it has the potential to become a hotspot.

[0121] Subsequently, the system continuously tracks the time windows following the candidate segment unit. When it detects that the actual hotspot scores of multiple consecutive candidate segment units (i.e., windows following the candidate segment) are all greater than the preset candidate threshold (or reach a more stringent confirmation threshold), the candidate segment unit is confirmed as a formal hotspot segment and included in the final hotspot segment list. Simultaneously, a fault-tolerance mechanism is introduced: if the hotspot score of subsequent windows fails to consistently exceed the threshold, the prediction is deemed a failure, and the candidate hotspot segment is removed from the hotspot queue to avoid generating invalid promotional materials.

[0122] In the above implementation process, by determining hotspot thresholds, merging continuous segments and correcting time sequence, and using a prediction and confirmation mechanism based on candidate thresholds, accurate extraction of hotspot segments, complete boundary definition and early prediction were achieved, effectively avoiding hotspot fragmentation and false alarms, and improving the timeliness and accuracy of hotspot discovery.

[0123] In this embodiment, step S210 is further included, aimed at generating promotional materials, including: extracting hot words and phrases from the bullet screen data group corresponding to the hot segments according to preset filtering conditions; generating a list of candidate hot words and phrases based on the hot words and phrases; and generating promotional materials for the competition-type live broadcast program corresponding to the multi-dimensional video data based on the program entity attention, program stage weight, and the list of candidate hot words and phrases. The promotional materials include at least one of the following: topic tags, cover text, and recommendation placement strategies corresponding to the competition-type live broadcast program. The specific implementation process is as follows:

[0124] First, from the bullet comment data group corresponding to the hot clips, hot word phrases that meet the requirements are screened out according to preset conditions. The screening conditions include that the phrase length is moderate, the occurrence count and the number of covered users reach the thresholds, the growth rate is relatively high, and the phrases contain evaluation words, exaggerated words, program entity words or emotion markers, etc., while excluding sensitive words, pure expressions or meaningless repeated characters. Subsequently, the screened hot word phrases are aggregated to form a hot word candidate phrase list, which is used as the core corpus for generating promotion materials. Then, based on the hot word candidate phrase list, the attention degree of program entities (such as the popularity of contestants or tracks) and the weight of program stages (such as the performance climax or voting stage), multiple types of promotion materials are automatically generated, including but not limited to:

[0125] Hashtag: Combine the hot word phrases with high-attention program entities (such as contestant names) to generate hashtags, and use the popularity of the entities to drive the spread of the topics.

[0126] Cover copy: Select the phrases with the strongest emotion or the most impact among the hot word phrases, and combine with program events to generate short copy and covers that attract clicks, so as to directly resonate with the audience's emotions.

[0127] Recommendation and delivery strategy: Intelligently determine the delivery time, channels (such as the short video public domain or the fan private domain), target groups and priority of the materials according to the hot spot score and the hot word score. For example, push the hot spots in the initial outbreak stage at the second level, and carry out precise targeted delivery to fans for the high-heat clips of specific contestants.

[0128] In this step, hot words are automatically screened and combined with program entities and stage features to generate hashtags, cover copies and delivery strategies, which realizes the intelligence and personalization of promotion materials, enables the materials to accurately carry the popularity of the live broadcast room, and significantly improves the promotion efficiency and user participation.

[0129] In the embodiment of the present application, a parameter feedback adjustment mechanism is further introduced after the promotion materials are delivered.

[0130] Specifically, after the promotion materials are delivered, multi-dimensional feedback data including the click volume, play completion rate, comment volume, forward volume, favorite volume, search volume and secondary creation volume are continuously collected to evaluate the actual communication effect of the materials. Based on the multi-dimensional feedback data, the core parameters are dynamically optimized through the feedback closed-loop mechanism: if the feedback of the materials corresponding to a certain type of hot spot feature (such as a specific weight combination or clustering radius) is good (such as high clicks and high play completion), the weight coefficient corresponding to this feature is automatically increased or the relevant screening threshold is relaxed; on the contrary, if the feedback of the materials is poor (such as low clicks and high bounce rate), the corresponding weight is reduced or the threshold is tightened. At the same time, the system will also adjust the hot spot threshold, prediction threshold, clustering radius adjustment parameter, similar group life cycle and attenuation weight according to the feedback, so that the algorithm can adapt to different program types (such as singing, dancing) and different communication environments, and ensure the continuous output of high-quality hot clips.

[0131] By dynamically adjusting the hotspot determination threshold, clustering parameters, and weight coefficients based on campaign feedback data, the solution achieves adaptive optimization, significantly improving the accuracy of hotspot identification and the dissemination effect of promotional materials.

[0132] In this embodiment, the limitations of traditional methods relying on manual monitoring or post-event statistics are overcome. A comprehensive processing scheme combining time window segmentation, dynamic clustering of bullet screen semantics, lifecycle decay mechanism, and proactive prediction of hotspot outbreaks is proposed. The overall process is as follows: Figure 3 As shown, multi-dimensional video data (including video timeline information, program flow information, bullet screen data, and user interaction data) of competitive live programs are acquired and divided into multiple candidate segment units according to a preset time window. The bullet screen data within the candidate segment units is subjected to text normalization, semantic vectorization, and dynamic clustering to generate similar groups (bullet screen data groups) with lifecycle and decay weights. Based on the similar groups, the first indicator data (data group concentration and hotspot outbreak prediction value) is determined, and combined with the second indicator data (program entity attention, program stage weight, bullet screen density growth value, user emotion intensity value, and user interaction growth value), a hotspot score is calculated by weighted fusion. When the hotspot score or hotspot outbreak prediction value meets the preset conditions, the corresponding candidate segment unit is determined as a hotspot segment or a candidate hotspot segment, and phrases that meet the screening conditions are extracted from the corresponding similar groups as hot word candidate phrases, further generating segment titles, topic tags, cover text, or recommendation placement strategies.

[0133] Overall, this application achieves a shift from passive response to proactive prediction by constructing an adaptive multi-dimensional hotspot score weighting mechanism that includes bullet screen density, similarity group concentration, emotional intensity, interaction growth, program stage, entity attention, and predicted value. This significantly reduces the delay in hotspot discovery and the cost of manual operation, effectively solves problems such as the numerous colloquial variations in bullet screen expression, significant differences in hotspot characteristics across different program stages, and interference from old hot words. It ensures a high degree of consistency between promotional materials (titles, tags, covers) and the real-time expression of the audience in the live broadcast room, thereby improving the accuracy of hotspot discovery, the automation level of material generation, and the overall dissemination efficiency of long-video competition live broadcast programs.

[0134] According to embodiments of this application, a video hotspot segment determination device is provided. It should be noted that the video hotspot segment determination device of this application embodiment can be used to execute the video hotspot segment determination method provided in this application embodiment. The following describes the video hotspot segment determination device provided in this application embodiment.

[0135] Figure 4 This is a structural diagram of a video hotspot segment determination device according to an embodiment of this application. Figure 4 As shown, the device includes:

[0136] The segmentation module 40 is used to segment the multidimensional video data to obtain multiple candidate segment units;

[0137] Clustering module 42 is used to cluster the bullet screen data within the candidate segment unit to obtain multiple bullet screen data groups, wherein the bullet screen data groups are used to represent a set of bullet screens that are semantically similar and located in the same time period.

[0138] The first determining module 44 is used to determine the first indicator data based on the barrage data group, and to determine the hot spot score based on the first indicator data and the second indicator data. The first indicator data is used to quantify the semantic consensus strength of the barrage and the hot spot outbreak trend, the second indicator data is used to quantify the hot spot contribution degree of the program context and the hot spot intensity of user interaction behavior, and the hot spot score is used to quantify the potential popularity of the candidate segment unit.

[0139] The second determining module 46 is used to determine hotspot segments from multiple candidate segment units based on hotspot scores.

[0140] By using the segmentation module, clustering module, first determination module, and second determination module in the aforementioned video hotspot segment determination device, the goal of locating high-popularity segments in live broadcast programs in real time and accurately is achieved. This results in improving the timeliness of hotspot discovery and optimizing the accuracy of hotspot judgment. Furthermore, it solves the technical problems of delayed hotspot recognition, fragmented bullet screen expressions, and difficulty in distinguishing the differentiated popularity characteristics of different program stages, which are common in competition-type long video live broadcast programs that usually rely on manual monitoring or post-event statistics to discover hotspots.

[0141] In the video hotspot segment determination device provided in this application embodiment, the segmentation module is further used to acquire multi-dimensional video data, wherein the multi-dimensional video data includes at least live video data, video timeline information, program flow information, bullet screen data, and user interaction data. The video timeline information includes at least the start and end times of the live broadcast of the competition-type live program, the start and end times of the segments, the real-time playback time, and the video frame timestamps. Based on the video timeline information, the live video data is divided into multiple candidate segment units according to a preset time window, and each candidate segment unit is associated with corresponding program flow information, bullet screen data, and user interaction data.

[0142] In the video hotspot segment determination device provided in this application embodiment, the clustering module is further used to perform text preprocessing on the bullet screen data within the candidate segment unit, and convert the preprocessed bullet screen data into bullet screen semantic vectors; determine the clustering radius based on the bullet screen density, bullet screen growth rate and the number of bullet screen data groups in the historical clustering results within the candidate segment unit, wherein the clustering radius is inversely proportional to the bullet screen density; perform clustering processing on the bullet screen semantic vectors based on the clustering radius to generate bullet screen data groups containing life cycle parameters and decay weights, wherein the life cycle parameter is used to represent the existence state of the bullet screen data group in the time series, and the decay weight is used to suppress the weight of bullet screen data groups that do not match the candidate segment unit.

[0143] In the video hotspot segment determination device provided in this application embodiment, the clustering module is further used to determine the number of times the bullet screen data group appears within a continuous time window; when the number of occurrences is greater than or equal to a preset number, the life cycle parameter is extended, and the attenuation weight is increased based on the number of occurrences, the number of covered users, and the bullet screen index growth rate, wherein the number of covered users is used to represent the number of users who have sent bullet screens within the bullet screen data group, and the bullet screen index growth rate is used to represent the rate of change of the bullet screen data group compared to the previous time window; when the number of occurrences is less than the preset number, or the bullet screen data group does not match the candidate segment unit, the attenuation weight is reduced according to a preset attenuation rule.

[0144] In the video hotspot segment determination device provided in this application embodiment, the first determination module is further used to determine the data group concentration degree based on the number of bullet comments in the largest bullet comment data group within the candidate segment unit and the total number of bullet comments within the candidate segment unit. The largest bullet comment data group is used to represent the bullet comment data group containing the most bullet comments within the candidate segment unit, and the data group concentration degree is used to reflect the semantic consistency degree of the bullet comment data within the candidate segment unit. A hotspot prediction model is used to analyze the data group concentration degree and the temporal feature sequence to obtain a hotspot outbreak prediction value. The temporal feature sequence includes at least one of the following: a sequence of bullet comment counts within multiple preset sliding windows, a sequence of user emotion intensity, a sequence of user interaction behavior, a sequence of program entities, and a sequence of program stage identifiers. The hotspot outbreak prediction value is used to represent the probability that the candidate segment unit will enter a hotspot state within a future time window.

[0145] In the video hotspot segment determination device provided in this application embodiment, the first determination module is further configured to match the candidate segment unit with the corresponding program entity and program stage based on the program flow information and bullet screen data within the candidate segment unit, to obtain a program entity association list and program stage identifier corresponding to the candidate segment unit; determine the program entity attention of the candidate segment unit based on the program entity association list, and determine the program stage weight of the candidate segment unit based on the program stage identifier, wherein the program entity attention is used to reflect the discussion heat of each program entity in the program entity association list, and the program stage weight is used to reflect the contribution of different program stages to the formation of hotspots; determine the bullet screen density growth value, user emotion intensity value, and user interaction growth value within the candidate segment unit; and determine the hotspot score of the candidate segment unit based on the data group concentration, hotspot outbreak prediction value, program entity attention, program stage weight, bullet screen density growth value, user emotion intensity value, and user interaction growth value.

[0146] In the video hotspot segment determination device provided in this application embodiment, the second determination module is further configured to determine the candidate segment unit as a hotspot segment when the hotspot score is greater than the preset hotspot threshold; when the hotspot scores of multiple consecutive candidate segment units are all greater than the preset hotspot threshold, the multiple candidate segment units are merged into a hotspot segment, and the start and end times of the hotspot segment are corrected according to the peak position of the hotspot scores of the multiple candidate segment units.

[0147] In the video hotspot segment determination device provided in this application embodiment, the second determination module is further configured to extract hot words and phrases from the barrage data group corresponding to the hotspot segment according to preset filtering conditions; generate a list of candidate hot words and phrases based on the hot words and phrases; and generate promotional materials for the competition-type live broadcast program corresponding to the multi-dimensional video data based on the program entity attention, program stage weight, and the list of candidate hot words and phrases. The promotional materials include at least one of the following: topic tags, cover text, and recommendation placement strategies corresponding to the competition-type live broadcast program.

[0148] This application also provides an electronic device, including: a memory and a processor, wherein the memory is used to store program instructions; and the processor is connected to the memory and used to execute the above-described video hotspot segment determination method.

[0149] It should be noted that the aforementioned electronic equipment is used to perform Figure 2 The method for determining video hotspot segments shown above also applies to this electronic device, and will not be repeated here.

[0150] This application also provides a non-volatile storage medium, which includes a stored computer program, wherein the device containing the non-volatile storage medium executes the above-described video hotspot segment determination method by running the computer program.

[0151] It should be noted that the aforementioned non-volatile storage media is used for execution. Figure 2 The video hotspot segment determination method shown above is also applicable to this non-volatile storage medium, and will not be repeated here.

[0152] This application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above-described method for determining video hotspot segments.

[0153] It should be noted that the above-mentioned computer program product is used to execute Figure 2 The method for determining video hotspot segments shown above also applies to this computer program product, and will not be repeated here.

[0154] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0155] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0156] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0157] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0158] Furthermore, the functional units in the various embodiments of this application 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.

[0159] 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, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0160] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for determining video hotspot segments, characterized in that, include: Multidimensional video data is divided into multiple candidate segment units; Clustering is performed on the bullet screen data within the candidate segment unit to obtain multiple bullet screen data groups; Based on the bullet screen data group, a first indicator data is determined, and a hot topic score is determined based on the first indicator data and the second indicator data. The first indicator data is used to quantify the semantic consensus strength of bullet screens and the hot topic outbreak trend, and the second indicator data is used to quantify the hot topic contribution degree of the program context and the hot topic intensity of user interaction behavior. The hot topic score is used to quantify the potential popularity of the candidate segment unit. Hotspot segments are determined from the plurality of candidate segment units based on the hotspot scores.

2. The method according to claim 1, characterized in that, The multidimensional video data is divided into multiple candidate segment units, including: Acquire multidimensional video data, wherein the multidimensional video data includes at least live video data, video timeline information, program flow information, bullet screen data, and user interaction data; Based on the video timeline information, the live video data is divided into multiple candidate segment units according to a preset time window, and each candidate segment unit is associated with corresponding program flow information, bullet screen data, and user interaction data.

3. The method according to claim 1, characterized in that, Clustering is performed on the bullet screen data within the candidate segment unit to obtain bullet screen data groups, including: The bullet screen data within the candidate segment unit is preprocessed with text, and the preprocessed bullet screen data is converted into bullet screen semantic vectors; The clustering radius is determined based on the bullet screen density, bullet screen growth rate, and the number of bullet screen data groups in the historical clustering results within the candidate segment unit. The danmaku semantic vectors are clustered based on the clustering radius to generate danmaku data groups containing lifecycle parameters and decay weights.

4. The method according to claim 3, characterized in that, The method further includes: Determine the number of times the bullet screen data group appears within a continuous time window; If the number of occurrences is greater than or equal to a preset number, the lifecycle parameter is extended, and the decay weight is increased based on the number of occurrences, the number of users covered, and the growth rate of the bullet screen metric. If the number of occurrences is less than the preset number, or if the bullet screen data group does not match the candidate segment unit, the attenuation weight is reduced according to the preset attenuation rule.

5. The method according to claim 1, characterized in that, The first indicator data includes data group concentration and hotspot outbreak prediction values. The first indicator data is determined based on the bullet screen data group, including: The concentration of the data group is determined based on the number of bullet comments in the largest bullet comment data group within the candidate segment unit and the total number of bullet comments within the candidate segment unit. The concentration of the data group is used to reflect the semantic consistency of the bullet comment data within the candidate segment unit. A hotspot prediction model is used to analyze the concentration and temporal characteristic sequence of the data group to obtain the hotspot outbreak prediction value, wherein the hotspot outbreak prediction value is used to represent the probability that the candidate segment unit will enter the hotspot state within a future time window.

6. The method according to claim 5, characterized in that, The second indicator data includes program entity attention, program stage weight, bullet screen density growth value, user emotion intensity value, and user interaction growth value. A hot topic score is determined based on the first and second indicator data, including: Based on the program flow information and bullet screen data within the candidate segment unit, the candidate segment unit is matched with the corresponding program entity and program stage to obtain the program entity association list and program stage identifier corresponding to the candidate segment unit. The program entity attention of the candidate segment unit is determined based on the program entity association list, and the program stage weight of the candidate segment unit is determined based on the program stage identifier. Determine the growth value of bullet screen density, user emotion intensity, and user interaction within the candidate segment unit; The hotspot score of the candidate segment unit is determined based on the data group concentration, the hotspot outbreak prediction value, the program entity attention value, the program stage weight, the bullet screen density growth value, the user emotion intensity value, and the user interaction growth value.

7. The method according to claim 6, characterized in that, Determining hotspot segments from the plurality of candidate segment units based on the hotspot score includes: If the hotspot score is greater than a preset hotspot threshold, the candidate segment unit is determined as a hotspot segment; When the hotspot scores of multiple consecutive candidate segment units are all greater than the preset hotspot threshold, the multiple candidate segment units are merged into a single hotspot segment, and the start and end times of the hotspot segment are adjusted according to the peak positions of the hotspot scores of the multiple candidate segment units.

8. The method according to claim 7, characterized in that, The method further includes: Hot words and phrases are extracted from the bullet screen data groups corresponding to the hot segments based on preset filtering conditions; A list of candidate hot words and phrases is generated based on the hot words and phrases. Based on the program entity attention, the program stage weight, and the list of hot word candidate phrases, promotional materials for the competition-type live program corresponding to the multi-dimensional video data are generated.

9. A device for determining video hotspot segments, characterized in that, include: The segmentation module is used to segment multidimensional video data to obtain multiple candidate segment units; The clustering module is used to perform clustering processing on the bullet screen data within the candidate segment unit to obtain multiple bullet screen data groups; The first determining module is used to determine first indicator data based on the bullet screen data group, and to determine a hot spot score based on the first indicator data and the second indicator data. The first indicator data is used to quantify the semantic consensus strength of bullet screens and the hot spot outbreak trend, the second indicator data is used to quantify the hot spot contribution degree of the program context and the hot spot intensity of user interaction behavior, and the hot spot score is used to quantify the potential popularity of the candidate segment unit. The second determining module is used to determine hotspot segments from the plurality of candidate segment units based on the hotspot score.

10. An electronic device, characterized in that, include: A memory and a processor, wherein the memory is used to store program instructions; The processor, connected to the memory, is used to execute the video hotspot segment determination method according to any one of claims 1 to 8.

Images