AI-based multimedia promotion content rapid generation method and system

Abstract

This invention provides an AI-based method and system for rapidly generating multimedia promotional content, belonging to the field of promotional content recommendation technology. By integrating multimodal user interaction data from text comments, audio features, and video viewing behavior, this invention achieves comprehensive analysis of user behavior, enabling the creation of more accurate user profiles. This generates personalized recommendation content, improving recommendation accuracy and better meeting users' actual needs, increasing user interest and acceptance of recommended content. By recognizing the emotional state exhibited by users during interactions, the invention can quickly respond to changes in user emotions and adjust recommendation content accordingly. An adaptive recommendation mechanism is introduced to continuously optimize recommendation strategies based on real-time changes in user emotional states. This allows for flexible adjustment of recommendation content in different contexts, providing promotional solutions that better meet users' current needs.

Description

Technical Field

[0001] The present invention relates to the technical field of promotional content recommendation, and specifically to a method and system for quickly generating multimedia promotional content based on AI. Background Art

[0002] With the popularity of social media and online platforms, users' interactive data in the Internet environment is becoming increasingly rich, including text comments, audio feedback, and video viewing records. These data not only reflect users' interests and preferences, but also provide important clues about their emotional state. In multimedia promotion scenarios, how to make full use of these rich data resources to quickly generate promotional content that meets user needs has become the key for companies to improve promotion effects and user engagement. This demand is particularly evident in many fields such as e-commerce, online education, entertainment, and social networks. Companies hope to attract more target users by optimizing promotion strategies.

[0003] At present, existing multimedia promotion technologies mainly rely on rule-based recommendation systems and simple content matching algorithms. These systems usually make recommendations based on users' historical behaviors, but lack in-depth understanding and analysis of users' emotional states. In addition, traditional methods mostly use static text matching technology and fail to effectively consider the characteristics of multimedia data such as audio and video. This results in that the promotion content often does not match the real needs of users, reducing the effectiveness of promotion and performing poorly in personalized recommendations.

[0004] Traditional multimedia promotion systems mainly rely on rule-based recommendation algorithms, which usually use users' historical behavior data for recommendations, such as purchase records and browsing history. However, these methods are mostly limited to the analysis of static data, and it is difficult to reflect users' emotional changes and dynamic interests in real time. Specifically, traditional recommendation systems often ignore the emotional characteristics of users in the process of interacting with content, and cannot deeply understand users' emotional responses to different types of content. In addition, the existing technology mainly uses text matching and keyword extraction in the recommendation of multimedia content, lacking comprehensive consideration of audio features and video viewing behavior. Existing solutions also have obvious deficiencies in personalized recommendations, usually focusing only on single-dimensional data and lacking multi-dimensional analysis of user behavior.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for quickly generating multimedia promotional content based on AI to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for quickly generating multimedia promotional content based on AI, the specific steps include:

[0009] Step 1: Collect user’s past Internet online interaction data within a certain period of time. The types of online interaction data collected include text comments posted by users, audio and video data viewed, in order to segment text comment data into separate words and remove stop words, obtain feature data of pitch and speech speed in audio data, obtain the number of videos viewed, the dwell time on each video and the average viewing time data;

[0010] Step 2: Extract keywords from the preprocessed text comment data to form a keyword set. Generate the user's real-time sentiment index based on the feature data of tone and speech speed in the audio data. Generate the user's recommended viewing time based on the dwell time and average viewing time data of each video viewed.

[0011] Step 3: Analyze each promotional video in the multimedia resource library, obtain the promotional text and video length data in each promotional video, calculate the text matching degree between the keywords and the promotional text in each promotional video, and perform sentiment scoring on the promotional text in each promotional video;

[0012] Step 4: Calculate the emotional matching index between each promotional video and the user based on the real-time emotional index and the promotional text, calculate the duration recommendation index of each promotional video based on the video length data of each promotional video and the recommended viewing time of the user, and combine the duration recommendation index with the number of videos viewed by the user to generate a recommended index;

[0013] Step 5: Construct a mathematical analysis model based on the text matching degree, sentiment matching index and suggestion recommendation index of each promotional video, screen the promotional videos in the multimedia resource library, and select the best promotional video in the multimedia resource library to push to users.

[0014] Furthermore, collect the user’s past When a user posts a text comment within a certain time period, All text comments within the time length are sorted by the time of posting of the comments and connected with specific commas to form a long string. When the text comment data is segmented into individual words and stop words are removed, the word segmentation tool is used to segment the words in the formed long string to generate a list containing all the words. The word segmentation results are traversed to filter and retain those words that are not in the stop word list to form a preprocessed word list;

[0015] The specific logic for obtaining the characteristic data of pitch and speech speed in audio data is as follows:

[0016] will pass All audio data within the time length are sorted and connected according to the release time, and the length between two adjacent audio data is The silent segments are connected, and the starting time point of the first audio data is set to 0, forming a complete audio segment and marked as , is the time variable;

[0017] Using Fourier transform Convert the time domain signal to the frequency domain signal to get the spectrum , in the spectrum Find the frequency component with the largest amplitude, i.e. the fundamental frequency , the fundamental frequency Marked as pitch data;

[0018] The audio signal is divided into several frames, and the time length of each frame is set to , the frame shift is set to , The total duration is , then the total number of frames for: Calculate the energy of each frame and compare it with the energy threshold. If the energy is greater than the energy threshold, , then the frame is marked as a speech frame, based on the formula: in, Indicates The energy of the frame, Indicates The energy start time of the frame, if , then the frame is marked as a speech frame, the number of all speech frames is counted, and the proportion of the total number of frames is calculated, and the proportion is marked as the speech rate data;

[0019] When obtaining the average viewing time data for the number of videos viewed, the user's stay time on each video is obtained. The formula for calculating the average viewing time is:

[0020] in, and Respectively indicate the user's past The total number of videos viewed and the average viewing time within the time period. Indicated in The duration of stay on the video.

[0021] Furthermore, the specific logic for determining the keywords that appear in the text is:

[0022] According to the preprocessed word list, calculate the frequency of each word in the word list, sort the frequency from high to low, and select the word with the highest frequency. words as keywords and form a keyword set;

[0023] The formula for generating the user's real-time sentiment index is: in, represents the average energy of the speech frame, Indicates The energy of a speech frame, Indicates the total number of speech frames, Represents the user's real-time sentiment index, Indicates the proportion of speech frames in the total number of frames.

[0024] Furthermore, the specific method for generating the recommended viewing time of the user is as follows:

[0025] For the dwell time and average viewing time data of each video browsed, calculate the ratio of the user's dwell time for each video to the average viewing time, based on the formula: in, Respectively represent the first The dwell time of the videos, and the ratio of dwell time to average viewing time, represents the average viewing time;

[0026] The weighted average dwell time is calculated based on the ratio of the user's dwell time on each video to the average viewing time. The recommended viewing time is obtained by combining the weighted average dwell time with the number of videos viewed by the user. The formula is: in, represents the weighted average residence time, Indicates the number of videos viewed by the user. Indicates the recommended viewing time.

[0027] Furthermore, the method for calculating the text matching degree between the keywords and the promotional text in each promotional video is as follows:

[0028] Divide the promotion text into individual words to form a promotion word set, find the intersection and union words of the promotion word set and the keyword set, and calculate the text matching degree. The formula is: in, Indicates keywords and The degree of text matching of the promotional text in the promotional video, and Respectively represent The number of intersection words and union words of the promotional word set and keyword set in the promotional video, Represents the keyword set and A collection of promotional words for a promotional video.

[0029] Furthermore, the method for calculating the sentiment matching index between each promotional video and the user is as follows:

[0030] Use the sentiment dictionary to perform sentiment scoring on the promotional text in each promotional video, obtain the sentiment score of the promotional text in each promotional video, and combine the user's real-time sentiment index generation and the user's sentiment matching index. The calculation formula is:

[0031] in, Respectively represent The sentiment score of the promotional text of the promotional video and the sentiment matching index between the real-time sentiment index of the user.

[0032] Furthermore, the formula for generating the recommendation index is:

[0033] in, Respectively represent The duration recommendation index and suggestion recommendation index of the promotional video. Respectively represents the recommended viewing time and the The duration of the promotional video, Indicates the number of videos viewed by the user.

[0034] Furthermore, a mathematical analysis model is constructed to screen various promotional videos in the multimedia resource library as follows:

[0035] The recommendation index of each promotional video is calculated based on the text matching degree, sentiment matching index and suggestion recommendation index of each promotional video. The formula is as follows:

[0036] in, Indicates The recommendation index of the promotional video;

[0037] The promotional videos are sorted in descending order according to the recommendation index, and the promotional video with the highest recommendation index is selected as the recommendation result and pushed to the user.

[0038] The present invention further provides an AI-based multimedia promotion content rapid generation system, the system is used to execute the above-mentioned multimedia promotion content rapid generation method, including:

[0039] Data collection module, used to collect user's past Internet online interaction data within a certain period of time. The types of online interaction data collected include text comments posted by users, audio and video data viewed, in order to segment text comment data into separate words and remove stop words, obtain feature data of pitch and speech speed in audio data, obtain the number of videos viewed, the dwell time on each video and the average viewing time data;

[0040] The preprocessing module is used to extract keywords from the text through the preprocessed text comment data to form a keyword set, generate the user's real-time emotion index based on the feature data of the tone and speech speed in the audio data, and generate the user's recommended viewing time based on the dwell time and average viewing time data of each video browsed;

[0041] The sentiment analysis module is used to analyze each promotional video in the multimedia resource library, obtain the promotional text and video length data in each promotional video, calculate the text matching degree between the keywords and the promotional text in each promotional video, and perform sentiment scoring on the promotional text in each promotional video;

[0042] The promotion analysis module is used to calculate the emotional matching index between each promotion video and the user based on the real-time emotional index and the promotion text, calculate the duration recommendation index of each promotion video based on the video length data of each promotion video and the recommended viewing time of the user, and combine the duration recommendation index with the number of videos browsed by the user to generate a recommendation index;

[0043] The video screening module is used to build a mathematical analysis model based on the text matching degree, sentiment matching index and suggestion recommendation index of each promotional video, screen the various promotional videos in the multimedia resource library, and select the best promotional video in the multimedia resource library to push to users.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention realizes a comprehensive analysis of user behavior by integrating multimodal user interaction data of text comments, audio features and video viewing behavior, and can establish a more accurate user portrait, thereby generating personalized recommended content, improving the accuracy of recommendations, and can better meet the actual needs of users, increase users' interest in and acceptance of recommended content, and can quickly respond to changes in user emotions and adjust recommended content in a timely manner by identifying the emotional state of users during the interaction process. It introduces an adaptive recommendation mechanism and continuously optimizes the recommendation strategy according to the real-time changes in the user's emotional state. It can flexibly adjust recommended content in different situations and provide promotion plans that better meet the current needs of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1It is a schematic diagram of the overall method flow of the present invention;

[0047] Figure 2 It is a schematic diagram of the overall system structure of the present invention. DETAILED DESCRIPTION

[0048] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0049] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0050] Example

[0051] See also Figure 1 , the present invention provides a technical solution:

[0052] A method for quickly generating multimedia promotional content based on AI, the specific steps include:

[0053] Step 1: Collect user’s past The types of online interactive data collected include text comments posted by users, audio and video data viewed, in order to segment the text comment data into separate words and remove stop words, obtain the feature data of pitch and speaking speed in the audio data, obtain the number of videos viewed, the dwell time on each video and the average viewing time data.

[0054] When collecting users' online Internet interaction data, it is necessary to implement a clear user notification and consent process, requiring users to read and agree to the privacy policy through pop-up windows or agreement boxes, and providing an optional consent button to ensure that users voluntarily participate in data collection with knowledge, ensure that users can easily withdraw their consent at any time, and inform them of the consequences of data processing after withdrawing their consent.

[0055] During the data collection and processing phase, the principle of data minimization must be followed. The operation steps include: for text comment data, using natural language processing techniques to segment the comments into individual words and remove stop words; for audio data, using audio analysis tools to extract features such as pitch and speech rate; for video browsing data, recording the number of videos browsed by the user, the dwell time of each video, and the average viewing time, without storing the user's identity information.

[0056] In this embodiment, when collecting the text comments posted by the user in the past time period, when sorting all the text comments in the past time period according to the posting time of the comments and connecting them with a specific comma to form a long string, when segmenting the text comment data into individual words and removing stop words, using the words in the long string formed by segmenting with a segmentation tool to generate a list containing all words, traversing the segmentation results, screening and retaining those words that are not in the stop word list to form a preprocessed word list.

[0057] When collecting the text comments of the user in the past time period, first, it is necessary to extract all the comments posted by the user during this time period from the database. Which database comments to extract specifically can be set according to requirements to ensure sorting by the posting time of the comments so that the order of the comments is retained in subsequent operations. Connect the sorted comment content with a specific comma to form a long string, which can ensure clear separation between each comment and prepare for subsequent word segmentation processing. Use a suitable programming language (such as Python) and data processing libraries (such as Pandas or NumPy) to implement data extraction and connection.

[0058] Using natural language processing libraries such as NLTK, spaCy, or jieba and other tools to segment the long string into individual words, the segmentation tool will automatically identify word boundaries according to language rules or machine learning models to generate a list containing all words. Establish a stop word list, which usually contains common words that are meaningless for text analysis (such as "的", "是", "在", etc.). Traverse the segmentation results, check each word one by one to see if it is in the stop word list, and screen and retain those words that are not in the stop words. The stop word list uses the default stop word list provided in the natural language processing library (such as NLTK, spaCy, jieba).

[0059] Furthermore, the specific logic for obtaining the feature data of pitch and speech rate in the audio data is as follows:

[0060] Sort and connect all the audio data in the past time period according to the posting time, and the length between two adjacent audio data is The silent segments are connected, and the starting time point of the first audio data is set to 0, forming a complete audio segment and marked as , As a time variable, it can ensure the continuity and fluency of the audio. By setting the start time of the first audio segment to 0, the entire audio signal The ability to perform analysis within a unified time frame makes subsequent feature extraction (such as pitch and speaking rate) more consistent, reduces errors caused by time inconsistencies, improves transitions between audio clips, avoids information loss, and thus improves the accuracy of the analysis.

[0061] Using Fourier transform Convert the time domain signal to the frequency domain signal to get the spectrum , in the spectrum Find the frequency component with the largest amplitude, i.e. the fundamental frequency , the fundamental frequency Marked as pitch data;

[0062] The audio signal is divided into several frames, and the time length of each frame is set to , the frame shift is set to , The total duration is , then the total number of frames for: Calculate the energy of each frame and compare it with the energy threshold. If the energy is greater than the energy threshold, , then the frame is marked as a speech frame, based on the formula: in, Indicates The energy of the frame, Indicates The energy start time of the frame, if , then the frame is marked as a speech frame, the number of all speech frames is counted, and the proportion of the total number of frames is calculated, and the proportion is marked as the speech rate data.

[0063] Fourier transform is used to transform the time domain signal Convert to frequency domain signal , which can efficiently extract the fundamental frequency in the audio signal As pitch data, Fourier transform can decompose complex audio signals into sinusoidal wave components of different frequencies, so that we can easily identify the main frequency components in the audio. The fundamental frequency usually represents the basic pitch characteristics of the audio. Then, the audio signal is divided into several frames and the energy of each frame is calculated to effectively identify the speech part. By setting the time length and frame shift of the frame, the audio signal can be locally analyzed to capture the dynamic changes of the audio signal. If the energy of a frame is greater than the preset energy threshold, it indicates that the frame contains a valid speech signal. This method can accurately distinguish between speech and non-speech parts, and then count the number of speech frames and calculate their proportion in the total number of frames. As a quantitative indicator of speech rate data, the energy threshold When making settings, the energy distribution of known speech frames and non-speech frames can be calculated based on historical audio data to make settings.

[0064] Through Fourier transform, we can quickly extract the pitch features in the audio signal, while frame energy analysis provides accurate recognition of speech activities. This combination makes the extraction of pitch and speech rate data more robust and reliable, and has strong adaptability. The use of this step can significantly improve the performance of the audio analysis system and provide more accurate basic data for subsequent audio recognition, sentiment analysis and other applications. This not only enhances the system's intelligent recognition capabilities, but also improves user experience, laying the foundation for the promotion and application of related technologies.

[0065] When obtaining the average viewing time data for the number of videos viewed, the user's stay time on each video is obtained. The formula for calculating the average viewing time is:

[0066] in, and Respectively indicate the user's past The total number of videos viewed and the average viewing time within the time period. Indicated in The duration of stay on the video.

[0067] Step 2: Extract keywords from the preprocessed text comment data to form a keyword set. Generate the user's real-time sentiment index based on the feature data of pitch and speech speed in the audio data. Generate the user's recommended viewing time based on the dwell time and average viewing time data of each video browsed.

[0068] The specific logic for determining the keywords that appear in the text is as follows: Based on the preprocessed word list, calculate the frequency of each word in the word list, sort the frequencies from high to low, and select the one with the highest frequency. words as keywords and form a keyword set.

[0069] Perform word segmentation on the original text, divide the text into individual words or phrases, count the preprocessed word list, count the frequency of each word in the list, and sort the words from high to low. You can use a sorting algorithm to achieve this function, select the word with the highest frequency from the sorted list The keywords are used to construct a keyword set. Through frequency statistics, important content and themes in the text can be quickly discovered. High-frequency words can often reflect the main focus and core topics of the text. Keywords can help simplify subsequent text analysis tasks. Through keyword extraction, the essence can be quickly extracted from a large amount of text, which is convenient for subsequent analysis and processing. Keywords can help quickly understand the main idea and content of the text, extract key points, and reduce reading and comprehension time. Keywords are used as input features for subsequent machine learning or natural language processing tasks to improve the effect and accuracy of the model. This embodiment can efficiently extract keywords from the text, which not only facilitates text analysis and processing, but also can intuitively reflect the theme and main information of the text.

[0070] The formula for generating the user's real-time sentiment index is: in, represents the average energy of the speech frame, Indicates The energy of a speech frame, Indicates the total number of speech frames, Represents the user's real-time sentiment index, Indicates the proportion of speech frames in the total number of frames.

[0071] By calculating the average energy of speech frames, we can understand the overall energy level of speech signals. In sentiment analysis, the intensity (energy) of speech can reflect the speaker's emotional state, such as excitement, anger, or calmness. When the emotion is strong, the energy of speech will increase accordingly. The fundamental frequency of the audio signal is usually related to the pitch of the speech. The change of fundamental frequency may reflect the change of emotional state. For example, high pitch is related to happiness or excitement, while low pitch may be related to sadness or indifference. The speech frame ratio is the proportion of speech frames in the total number of frames, which is used to measure the activity of speech. A higher ratio indicates the speaker's confidence and emotional involvement in expressing emotions. High-energy speech is usually associated with strong emotions, reflecting the speaker's emotional state. Therefore, the average energy affects the calculation of the sentiment index through logarithmic transformation, making the expression of emotion more significant at relatively high energy.

[0072] Fundamental frequency is directly related to intonation, and changes in intonation play an important role in emotional expression. Higher fundamental frequencies are associated with positive emotions, while lower fundamental frequencies may indicate negative emotions. Therefore, taking the square root as a weight can emphasize the impact of frequency on emotion. The voice share factor takes into account the degree of voice activity. Active voice expression usually leads to clearer emotional communication. The more frequent the voice, the higher the speaker's investment and involvement in emotional expression, thus affecting the emotional index. A higher emotional index indicates that the speaker may be in a more excited, positive or strong emotional state. For example, emotions such as happiness, excitement or anger may result in higher values. Conversely, a lower emotional index may reflect a neutral or negative emotional state, such as boredom, sadness or indifference.

[0073] Logarithmic transformations are often used to compress the range of values, especially when dealing with exponentially growing phenomena. The intensity of emotional expressions is not linearly related, and using logarithms can make small changes more significant in the low energy range and smoother in the high energy range, which helps avoid inadvertently amplifying extremely high energy values, making the emotional index more stable.

[0074] A higher voice frame ratio means that the speaker is more engaged and active in expressing emotions, thus increasing their influence in the emotional index, which reflects that the frequency of emotional expression is proportional to the intensity of emotion to some extent. The fundamental frequency is directly related to the pitch of the voice, and changes in pitch are usually accompanied by changes in emotions. High fundamental frequency is usually associated with positive emotions (such as happiness and excitement), while low fundamental frequency may be associated with negative emotions (such as sadness and indifference). By taking the square root of the fundamental frequency, its contribution to the emotional index can be emphasized, reflecting the impact of changes in pitch on emotional expression, and combining multiple factors that affect emotions, so that the final emotional index can more comprehensively reflect the speaker's emotional state.

[0075] The specific method for generating the recommended viewing time for a user is:

[0076] For the dwell time and average viewing time data of each video browsed, calculate the ratio of the user's dwell time for each video to the average viewing time, based on the formula: in, and represent the first The dwell time of the videos, and the ratio of dwell time to average viewing time, represents the average viewing time;

[0077] The weighted average dwell time is calculated based on the ratio of the user's dwell time on each video to the average viewing time. The recommended viewing time is obtained by combining the weighted average dwell time with the number of videos viewed by the user. The formula is: in, represents the weighted average residence time, Indicates the number of videos viewed by the user. Indicates the recommended viewing time.

[0078] The specific method of generating the recommended viewing time for users is based on the ratio of dwell time to average viewing time, which reflects the user's interest and engagement in different videos. By calculating the ratio of dwell time to average viewing time for each video, the user's interest in the video can be quantified. If the ratio is greater than 1, it means that the user's dwell time on the video exceeds the average viewing time, which may indicate a high interest in the video. On the contrary, if it is less than 1, it means that the user's interest may be low.

[0079] The weighted average dwell time combines the dwell time of each video and its corresponding ratio, which ensures that videos with longer user dwell time and higher ratio have a greater weight in the calculation. This weighting mechanism can more accurately reflect the user's preference for video content. By dividing the weighted average dwell time by the number of videos viewed by the user, a recommended viewing time suitable for the user can be calculated. This process ensures that the recommended viewing time matches the user's actual dwell behavior.

[0080] By calculating the ratio of dwell time to average viewing time, the user's interest in each video content can be captured more effectively. This enables the recommendation system to better meet user preferences, and the weighted average calculation method avoids the excessive impact of extreme values ​​of a single video's dwell time on the overall recommendation duration. Only when the user's dwell time on a video is significantly higher than the average level will the video have a greater impact on the recommendation results. The recommended viewing time can be dynamically adjusted based on the user's actual viewing behavior, rather than statically setting a fixed duration. This flexibility can improve the relevance of recommendations and user satisfaction. By combining dwell time with average viewing time, it is possible to analyze user engagement with video content from a quantitative perspective, thereby providing users with more personalized viewing recommendations.

[0081] The method for calculating the recommended viewing time in this embodiment analyzes the ratio of the user's dwell time on different videos to the average viewing time, and combines it with a weighted average method, so that the final recommended viewing time can reflect the user's preferences while avoiding interference from extreme values.

[0082] Step 3: Analyze each promotional video in the multimedia resource library, obtain the promotional text and video length data in each promotional video, calculate the text matching degree between the keywords and the promotional text in each promotional video, and perform sentiment scoring on the promotional text in each promotional video.

[0083] In this embodiment, the method for calculating the text matching degree between the keyword and the promotional text in each promotional video is:

[0084] Divide the promotion text into individual words to form a promotion word set, find the intersection and union words of the promotion word set and the keyword set, and calculate the text matching degree. The formula is: in, Indicates keywords and The degree of text matching of the promotional text in the promotional video, and Respectively represent The number of intersection words and union words of the promotional word set and keyword set in the promotional video, Represents the keyword set and A collection of promotional words for a promotional video.

[0085] The formula for calculating the degree of text matching aims to evaluate the relevance and consistency of the promotional content by combining and comparing the set of keywords and the set of promotional texts. This process is crucial to understanding the appeal and effectiveness of promotional videos. In text matching, set theory provides a clear way to quantify the overlap and diversity of vocabulary. The similarity between keywords and promotional texts can be intuitively understood through the concepts of intersection and union. The intersection reflects the direct connection between keywords and promotional texts, that is, how many keywords are included in the promotional text. A higher number of intersections means that the text content is closer to the core theme or information. The union represents the number of all unique words, helping to establish a benchmark that indicates the diversity of keywords and promotional texts. The larger this value, the richer the content. By calculating the ratio of intersection to union, a matching degree ranging from 0 to 1 can be obtained. When the text matching degree is greater, it means that the promotional text is completely composed of keywords, indicating a high degree of relevance, which means that the promotional content can effectively attract the target audience. At this point, users are more likely to be interested in the promotional video and watch or interact further. When the text matching degree is smaller, it means that the promotional text does not contain any keywords, indicating that the relevance is extremely low, indicating that the correlation between the promotional text and the keywords is low, which may mean that the content fails to effectively convey the theme or information, thereby reducing user interest and affecting the promotion effect.

[0086] Keywords are the core of promotional content and are directly related to user interest and understanding. When keywords are included in promotional text, it indicates that the content is more consistent with user needs or preferences. By comparing keywords and promotional text, the consistency of promotional content can be evaluated. A high degree of match means that the promotional text clearly conveys the information represented by the keywords, thereby enhancing the effectiveness of the promotion. The degree of text matching can not only affect the user's emotional response, but may also have an impact on behavioral indicators such as click-through rate and viewing time. Users are more likely to be interested in content that matches their search intent. The calculation method of text matching provides an effective way to evaluate the relevance and effectiveness of content by analyzing the relationship between keywords and promotional text. The use of intersection and union combined with the intuitiveness of set theory makes this evaluation process simple and powerful. In this way, the effect of promotional videos can be quantified, providing a basis for optimizing content and improving user experience.

[0087] Step 4: Calculate the emotional matching index between each promotional video and the user based on the real-time emotional index and the promotional text, calculate the duration recommendation index of each promotional video based on the video length data of each promotional video and the user's recommended viewing time, and combine the duration recommendation index with the number of videos viewed by the user to generate a recommended index.

[0088] The real-time sentiment index can adjust the recommended content according to the user's current emotional state to provide a more personalized viewing experience. This dynamic adjustment can improve the resonance between users and content and enhance user engagement. Considering the length of the video and the user's recommended viewing time, it can ensure that the length of the recommended video matches the user's attention and time allocation. This combination helps to increase the probability of users completing the viewing and reduce the number of users who give up midway due to the video being too long. Through sentiment scoring and text matching, it can ensure that the recommended content is not only consistent with the user's preferences, but also consistent with the user's current emotional state. This can improve user viewing satisfaction and enhance the attractiveness and relevance of the content.

[0089] In this embodiment, the method for calculating the emotion matching index between each promotional video and the user is as follows:

[0090] Use the sentiment dictionary to perform sentiment scoring on the promotional text in each promotional video, obtain the sentiment score of the promotional text in each promotional video, and combine the user's real-time sentiment index generation and the user's sentiment matching index. The calculation formula is: in, Respectively represent The sentiment score of the promotional text of the promotional video and the sentiment matching index between the real-time sentiment index of the user.

[0091] The sentiment score of the promotional text of the promotional video is obtained by analyzing the text through the sentiment dictionary. The sentiment dictionary contains a series of sentiment words and their corresponding sentiment polarity (positive, negative or neutral) and strength. When the system performs sentiment analysis on the promotional text, it will identify the sentiment words contained in the text and calculate a comprehensive sentiment score based on the sentiment polarity and strength of the words.

[0092] The sentiment matching index represents the sentiment matching index between the promotion video and the user's real-time sentiment index, which is obtained by combining the sentiment score of the current promotion video and the user's real-time sentiment index. represents the product of the promoted video sentiment score and the user's sentiment state, reflecting the correlation between the two; and It is used for normalization to ensure that the value of the sentiment matching index is within a reasonable range, so as to balance the influence of the sentiment score and user sentiment.

[0093] The sentiment matching index can intuitively reflect the degree of fit between the video content and the user's emotional state. When the sentiment matching index value is high, it means that the emotional content of the promotional video is consistent with the user's emotional state, which may arouse the user's interest and resonance, thereby increasing the user's willingness to watch and interaction rate. The sentiment dictionary is a basic tool for sentiment analysis, which is usually manually constructed by researchers or developers or trained based on existing data sets. The dictionary contains a large number of sentiment words, which are marked as positive, negative or neutral, and accompanied by corresponding sentiment intensity values. For example, "happy" may be marked as positive with an intensity of 2, while "sad" is marked as negative with an intensity of 3. When scoring the text of each promotional video, the system first segments the text and extracts the words in it. Then, the system compares these words with the words in the sentiment dictionary to find the sentiment words contained in the text. For each sentiment word, according to its sentiment polarity and intensity, the system will accumulate the score to form the overall sentiment score of the text.

[0094] The formula for generating the recommended index is: in, Respectively represent The duration recommendation index and suggestion recommendation index of the promotional video. Respectively represents the recommended viewing time and the The duration of the promotional video, Indicates the number of videos viewed by the user.

[0095] The duration recommendation index is obtained by comparing the user's recommended viewing time with the video duration. The duration recommendation index reflects whether the video is suitable for the user's viewing time. If the duration recommendation index value is high, it means that the video length matches the user's recommended viewing time, which may increase the user's willingness to watch. By multiplying the duration recommendation index with the number of videos viewed by the user, it is suggested that the recommendation index can more comprehensively reflect the user's viewing habits and behaviors. If the number of videos viewed by the user is large, it means that they are more receptive to the video content, so the recommendation index of the video will also increase accordingly.

[0096] The recommendation effect is optimized by matching the duration and quantifying the user's behavior. The duration recommendation index ensures that the recommended content matches the user's viewing time, while the suggestion recommendation index takes into account the user's active viewing behavior, forming a more comprehensive and dynamic recommendation mechanism. This method can improve the user's viewing experience and increase the user's satisfaction with the recommended content, thereby enhancing the platform's user stickiness and content consumption rate.

[0097] The larger the recommendation index, the more closely the promoted video matches the user's viewing habits and recommended viewing duration. A larger recommendation index means that the video duration is relatively close to the user's recommended viewing duration, and the user is more likely to choose to watch the video within a limited time. Users may feel that this video is more suitable within their schedule, thereby increasing their willingness to watch. Since the recommendation index also takes into account the number of videos viewed by the user, a large recommendation index also implies that the user has shown a high level of interest and engagement in similar content in their past viewing behavior. Therefore, the system believes that this video is more attractive in terms of the user's viewing preferences.

[0098] Step 5: Construct a mathematical analysis model based on the text matching degree, sentiment matching index and suggestion recommendation index of each promotional video, screen the promotional videos in the multimedia resource library, and select the best promotional video in the multimedia resource library to push to users.

[0099] Construct a mathematical analysis model to screen the promotional videos in the multimedia resource library:

[0100] The recommendation index of each promotional video is calculated based on the text matching degree, sentiment matching index and suggestion recommendation index of each promotional video. The formula is as follows:

[0101] in, Indicates The recommendation index of the promotional video;

[0102] The promotional videos are sorted in descending order according to the recommendation index, and the promotional video with the highest recommendation index is selected as the recommendation result and pushed to the user.

[0103] Indicates The recommendation index of a promotional video is calculated based on the text matching degree, sentiment matching index and suggestion recommendation index. It uses a combination of multiple factors and emphasizes the importance of text matching through an exponential function to ensure that the relevance of the text has an exponential impact. At the same time, the logarithmic form of the sentiment matching index ensures that the impact of sentiment matching is gradual, avoiding extreme values ​​from having too much impact on the recommendation index. Finally, the suggestion recommendation index increases its contribution to the overall index through the square root form while maintaining a positive value, thereby enhancing the recommendation effect.

[0104] The recommendation index can comprehensively reflect the multi-dimensional characteristics of each video to ensure the accuracy and effectiveness of the recommendation. The text matching degree reflects the fit between the video content and user needs, the emotional matching index reflects the resonance between the user's emotional tendency and the video content, and the recommendation index takes into account the user's viewing habits and the adaptability of the video. This combination enables the model to evaluate the recommendation value of each video from multiple aspects.

[0105] The correlation between the recommendation index and the internal variables (text match, sentiment match, and suggestion index) is that they together constitute the user's overall perception and expectations of video content. By calculating these variables, the model can identify the video that best matches the user's needs, ensuring that the final recommended content not only meets the user's emotional needs, but also has appropriateness in length and subject matter. This recommendation mechanism based on multivariate analysis aims to improve the user experience and increase user satisfaction with the recommended content, thereby achieving higher viewing and interaction rates.

[0106] The text matching degree reflects the relevance of the video content to the user's interests or needs. The higher the relevance, the higher the user's acceptance and click-through rate of the video. The emotional matching index quantifies the emotional appeal of the video content and reflects the degree of emotional resonance of the user when watching the video. Emotional resonance can enhance the user's viewing experience. The recommendation index measures the suitability of the video, combining the video length with the user's viewing habits. It ensures that users can take their own disposable time into account when recommending videos.

[0107] The larger the value of the recommendation index, the more the video fits the user's needs and preferences in the above dimensions. The exponential function ensures that the impact of text matching is nonlinear, that is, a small change in correlation can cause a significant change in the recommendation level, emphasizing the importance of text content. The logarithmic function can suppress the excessive impact of extreme sentiment values ​​on the results, ensuring that the model is more robust. By combining the square root form , the model can better adapt to the viewing habits of different users and increase the overall recommendation accuracy.

[0108] Sorting the promoted videos according to the recommendation index and selecting the video with the highest recommendation index as the recommendation result can significantly improve the accuracy of the recommendation and the user experience. By comprehensively considering the text matching degree, sentiment matching index and suggestion recommendation index, this method can better meet the personalized needs of users, thereby increasing the click-through rate and viewing rate. In addition, quickly screening out the videos that best suit the user's interests can optimize the decision-making process, reduce the time users spend searching for content, and thus reduce the user churn rate.

[0109] Traditional recommendation systems often rely on a single factor for recommendation, but this embodiment forms a comprehensive evaluation through multi-dimensional analysis, which improves the intelligence level of the system. Personalized recommendations not only increase user participation, but also promote the platform's data accumulation and feedback mechanism, forming a virtuous circle, and improving the effectiveness and competitiveness of the recommendation system as a whole.

[0110] This embodiment achieves a comprehensive analysis of user behavior by integrating multimodal user interaction data such as text comments, audio features, and video viewing behavior, and can establish a more accurate user portrait, thereby generating personalized recommended content, improving the accuracy of recommendations, and better meeting the actual needs of users, increasing users' interest in and acceptance of recommended content. By identifying the emotional state of users during the interaction process, it can quickly respond to changes in user emotions and adjust recommended content in a timely manner. For example, when a user shows positive emotions towards a certain type of content, the system can give priority to recommending similar content; otherwise, the push of such content can be reduced. This dynamic feedback mechanism not only enhances the user experience, but also effectively improves the relevance of content and user engagement.

[0111] This embodiment introduces an adaptive recommendation mechanism to continuously optimize the recommendation strategy based on the real-time changes in the user's emotional state. This technical feature enables the system to flexibly adjust the recommended content in different situations and provide a promotion plan that better meets the user's current needs.

[0112] See also Figure 2 The present invention also provides an AI-based multimedia promotion content rapid generation system, the system is used to execute the above-mentioned multimedia promotion content rapid generation method, including:

[0113] Data collection module, used to collect user's past Internet online interaction data within a certain period of time. The types of online interaction data collected include text comments posted by users, audio and video data viewed, in order to segment text comment data into separate words and remove stop words, obtain feature data of pitch and speech speed in audio data, obtain the number of videos viewed, the dwell time on each video and the average viewing time data;

[0114] The preprocessing module is used to extract keywords from the text through the preprocessed text comment data to form a keyword set, generate the user's real-time emotion index based on the feature data of the tone and speech speed in the audio data, and generate the user's recommended viewing time based on the dwell time and average viewing time data of each video browsed;

[0115] The sentiment analysis module is used to analyze each promotional video in the multimedia resource library, obtain the promotional text and video length data in each promotional video, calculate the text matching degree between the keywords and the promotional text in each promotional video, and perform sentiment scoring on the promotional text in each promotional video;

[0116] The promotion analysis module is used to calculate the emotional matching index between each promotion video and the user based on the real-time emotional index and the promotion text, calculate the duration recommendation index of each promotion video based on the video length data of each promotion video and the recommended viewing time of the user, and combine the duration recommendation index with the number of videos browsed by the user to generate a recommendation index;

[0117] The video screening module is used to build a mathematical analysis model based on the text matching degree, sentiment matching index and suggestion recommendation index of each promotional video, screen the various promotional videos in the multimedia resource library, and select the best promotional video in the multimedia resource library to push to users.

[0118] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0119] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. Those skilled in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0120] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0121] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A method for quickly generating multimedia promotional content based on AI, characterized in that: The specific steps include: Step 1: Collect user’s past Internet online interaction data within a certain period of time. The types of online interaction data collected include text comments posted by users, audio and video data viewed, in order to segment text comment data into separate words and remove stop words, obtain feature data of pitch and speech speed in audio data, obtain the number of videos viewed, the dwell time on each video and the average viewing time data; Step 2: Extract keywords from the preprocessed text comment data to form a keyword set. Generate the user's real-time sentiment index based on the feature data of tone and speech speed in the audio data. Generate the user's recommended viewing time based on the dwell time and average viewing time data of each video viewed. Step 3: Analyze each promotional video in the multimedia resource library, obtain the promotional text and video length data in each promotional video, calculate the text matching degree between the keywords and the promotional text in each promotional video, and perform sentiment scoring on the promotional text in each promotional video; Step 4: Calculate the emotional matching index between each promotional video and the user based on the real-time emotional index and the promotional text, calculate the duration recommendation index of each promotional video based on the video length data of each promotional video and the recommended viewing time of the user, and combine the duration recommendation index with the number of videos viewed by the user to generate a recommended index; Step 5: Construct a mathematical analysis model based on the text matching degree, sentiment matching index and suggestion recommendation index of each promotional video, screen the promotional videos in the multimedia resource library, and select the best promotional video in the multimedia resource library to push to users.

2. The method for rapidly generating multimedia promotional content based on AI according to claim 1, characterized in that: Collect user past When a user posts a text comment within a certain time period, All text comments within the time length are sorted by the time of posting of the comments and connected with specific commas to form a long string. When the text comment data is segmented into individual words and stop words are removed, the word segmentation tool is used to segment the words in the formed long string to generate a list containing all the words. The word segmentation results are traversed to filter and retain those words that are not in the stop word list to form a preprocessed word list; The specific logic for obtaining the characteristic data of pitch and speech speed in audio data is as follows: will pass All audio data within the time length are sorted and connected according to the release time, and the length between two adjacent audio data is The silent segments are connected, and the starting time point of the first audio data is set to 0, forming a complete audio segment and marked as , is the time variable; Using Fourier transform Convert the time domain signal to the frequency domain signal to get the spectrum , in the spectrum Find the frequency component with the largest amplitude, i.e. the fundamental frequency , the fundamental frequency Marked as pitch data; The audio signal is divided into several frames, and the time length of each frame is set to , the frame shift is set to , The total duration is , then the total number of frames for: Calculate the energy of each frame and compare it with the energy threshold. If the energy is greater than the energy threshold, , then the frame is marked as a speech frame, based on the formula: in, Indicates The energy of the frame, Indicates The energy start time of the frame, if , then the frame is marked as a speech frame, the number of all speech frames is counted, and the proportion of the total number of frames is calculated, and the proportion is marked as the speech rate data; When obtaining the average viewing time data for the number of videos viewed, the user's stay time on each video is obtained. The formula for calculating the average viewing time is: in, and Respectively indicate the user's past The total number of videos viewed and the average viewing time within the time period. Indicated in The duration of stay on the video.

3. The method for rapidly generating multimedia promotional content based on AI according to claim 2, characterized in that: The specific logic for determining the keywords that appear in the text is: According to the preprocessed word list, calculate the frequency of each word in the word list, sort the frequency from high to low, and select the word with the highest frequency. words as keywords and form a keyword set; The formula for generating the user's real-time sentiment index is: in, represents the average energy of the speech frame, Indicates The energy of a speech frame, Indicates the total number of speech frames, Represents the user's real-time sentiment index, Indicates the proportion of speech frames in the total number of frames.

4. The method for rapidly generating multimedia promotional content based on AI according to claim 3, characterized in that: The specific method for generating the recommended viewing time for a user is: For the dwell time and average viewing time data of each video browsed, calculate the ratio of the user's dwell time for each video to the average viewing time, based on the formula: in, and represent the first The dwell time of the videos, and the ratio of dwell time to average viewing time, represents the average viewing time; The weighted average dwell time is calculated based on the ratio of the user's dwell time on each video to the average viewing time. The recommended viewing time is obtained by combining the weighted average dwell time with the number of videos viewed by the user. The formula is: in, represents the weighted average residence time, Indicates the number of videos viewed by the user. Indicates the recommended viewing time.

5. The method for rapidly generating multimedia promotional content based on AI according to claim 1, characterized in that: The method for calculating the text matching degree between the keywords and the promotional text in each promotional video is as follows: Divide the promotion text into individual words to form a promotion word set, find the intersection and union words of the promotion word set and the keyword set, and calculate the text matching degree. The formula is: in, Indicates keywords and The degree of text matching of the promotional text in the promotional video, and Respectively represent The number of intersection words and union words of the promotional word set and keyword set in the promotional video, Represents the keyword set and A collection of promotional words for a promotional video.

6. The method for rapidly generating multimedia promotional content based on AI according to claim 5, characterized in that: The method for calculating the sentiment matching index between each promotional video and the user is: Use the sentiment dictionary to perform sentiment scoring on the promotional text in each promotional video, obtain the sentiment score of the promotional text in each promotional video, and combine the user's real-time sentiment index generation and the user's sentiment matching index. The calculation formula is: in, Respectively represent The sentiment score of the promotional text of the promotional video and the sentiment matching index between the real-time sentiment index of the user.

7. The method for rapidly generating multimedia promotional content based on AI according to claim 6, characterized in that: The formula for generating the recommended index is: in, Respectively represent The duration recommendation index and suggestion recommendation index of the promotional video. Respectively represents the recommended viewing time and the The duration of the promotional video, Indicates the number of videos viewed by the user.

8. The method for rapidly generating multimedia promotional content based on AI according to claim 7, characterized in that: Construct a mathematical analysis model to screen the promotional videos in the multimedia resource library: The recommendation index of each promotional video is calculated based on the text matching degree, sentiment matching index and suggestion recommendation index of each promotional video. The formula is as follows: in, Indicates The recommendation index of the promotional video; The promotional videos are sorted in descending order according to the recommendation index, and the promotional video with the highest recommendation index is selected as the recommendation result and pushed to the user.

9. An AI-based multimedia promotion content rapid generation system, characterized by: The system is used to execute a method for quickly generating multimedia promotion content according to any one of claims 1 to 8, comprising: Data collection module, used to collect user's past Internet online interaction data within a certain period of time. The types of online interaction data collected include text comments posted by users, audio and video data viewed, in order to segment text comment data into separate words and remove stop words, obtain feature data of pitch and speech speed in audio data, obtain the number of videos viewed, the dwell time on each video and the average viewing time data; The preprocessing module is used to extract keywords from the text through the preprocessed text comment data to form a keyword set, generate the user's real-time emotion index based on the feature data of the tone and speech speed in the audio data, and generate the user's recommended viewing time based on the dwell time and average viewing time data of each video browsed; The sentiment analysis module is used to analyze each promotional video in the multimedia resource library, obtain the promotional text and video length data in each promotional video, calculate the text matching degree between the keywords and the promotional text in each promotional video, and perform sentiment scoring on the promotional text in each promotional video; The promotion analysis module is used to calculate the emotional matching index between each promotion video and the user based on the real-time emotional index and the promotion text, calculate the duration recommendation index of each promotion video based on the video length data of each promotion video and the recommended viewing time of the user, and combine the duration recommendation index with the number of videos browsed by the user to generate a recommendation index; The video screening module is used to build a mathematical analysis model based on the text matching degree, sentiment matching index and suggestion recommendation index of each promotional video, screen the various promotional videos in the multimedia resource library, and select the best promotional video in the multimedia resource library to push to users.

Images