Live broadcast script intelligent optimization system and method

Abstract

The invention relates to the technical field of computers, in particular to a live broadcast script intelligent optimization system and method, and the system comprises a multi-source data collection module which is used for collecting audience data, live broadcast script data and historical feedback data; the script optimization engine comprises a content generation unit which generates a script framework based on the LSTM neural network according to the live broadcast theme, the commodity information and the target audience characteristics; the quality evaluation unit is used for carrying out quantitative scoring on the scripts through a multi-index fusion model; the dynamic optimization unit is used for adjusting script content and rhythm based on a reinforcement learning algorithm in combination with audience real-time feedback; the multi-dimensional analysis module is used for audience portrait analysis, creativity popularity prediction and commercial value evaluation; and the real-time feedback system receives audience feedback in live broadcast and drives the script optimization engine to perform dynamic adjustment. The efficiency is improved, the quality is optimized, and an industrial unified script quality evaluation system is established through standardized output.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a live streaming script intelligent optimization system and method. Background Technology

[0002] With the explosive growth of the live streaming industry, live streaming scripts have become a core element in ensuring the effectiveness of live streams. An existing invention, CN117750114A, entitled "A Method and System for Generating Live Stream Script Text," utilizes real-time script effectiveness calculations based on the text bias and response dispersion of different bullet comments. This effectively quantifies the volatility of bullet comments caused by inappropriate comments during the live stream, as well as the trends in audience feedback based on the live stream content and text, enhancing the sensitivity of identifying bullet comment emotions. This plays a practical role in increasing the attractiveness of live streams, boosting live stream traffic, and improving the quality of live streaming script text content. However, it lacks data utilization: it lacks in-depth mining of key data such as audience behavior, interaction data, and commercial conversion. Summary of the Invention

[0003] This invention provides a live streaming script intelligent optimization system that integrates data-driven analysis, intelligent algorithm optimization, and real-time feedback mechanisms. It solves problems such as low manual efficiency, inconsistent optimization standards, and insufficient data utilization, thereby improving script quality and the commercial value of live streaming.

[0004] This invention provides a live streaming script intelligent optimization system, the architecture of which includes: The multi-source data acquisition module is used to collect audience data, live stream script data, and historical feedback data; among them, Viewer data collected includes region, age, gender, viewing duration, number of interactions, and spending amount; live stream script data includes theme, duration, and number of products; historical feedback data includes feedback type, feedback content, feedback time, processing status, satisfaction level, and reasons for annotation. Script optimization engine, including: The content generation unit generates a script framework based on the LSTM neural network according to the live broadcast topic, product information, and target audience characteristics. Applications of LSTM networks in content generation include: Data preprocessing: converting text data such as live stream topics, product information, and target audience characteristics into word vector representations, for example, using Word2Vec or GloVe models; Model training: training the LSTM network using labeled live stream script data to learn the grammatical, semantic, and structural features in the text sequence; Script generation: based on the input live stream topics, product information, and target audience characteristics, the LSTM network outputs the next most likely word or phrase, gradually building a complete script framework.

[0005] The quality assessment unit uses a multi-indicator fusion model to quantitatively score the scripts; the multi-indicator fusion model includes a legal accuracy model, a content relevance model, and a pacing rationality model to quantitatively score the scripts.

[0006] The dynamic optimization unit adjusts the script content and pacing based on reinforcement learning algorithms and real-time audience feedback; real-time audience feedback includes sentiment analysis of bullet comments and dwell time, etc. The multi-dimensional analysis module is used for audience profiling, creative popularity prediction, and commercial value assessment. Audience profiling analysis: Based on clustering algorithms, user segmentation models are constructed, and personalized script suggestions are output; Creative popularity prediction: Predict the popularity trend of creative keywords through time series models, such as the ARIMA model; Business value assessment: Establish a regression model between product conversion rate and script elements, where the script elements are promotional message density and interaction frequency.

[0007] The real-time feedback system receives audience feedback during the live stream and drives the script optimization engine to make dynamic adjustments. Audience feedback includes bullet comments, likes, and purchasing behavior. The system drives the script optimization engine to adjust the live stream segments in real time, and makes dynamic adjustments such as increasing the frequency of interaction and adjusting the order of product display.

[0008] Preferably, the multi-source data acquisition module includes: The audience data collection submodule collects audience data such as region, age, gender, viewing duration, number of interactions, and spending amount. The script data collection submodule collects information such as the live stream topic, script duration, number of products, and estimated sales revenue. The feedback data collection submodule collects viewer comments, likes, purchase behavior, and satisfaction ratings.

[0009] Preferably, the multi-index fusion model of the quality assessment unit includes: Syntax accuracy metrics, based on NLP syntax tree analysis; Content relevance metrics are calculated using cosine similarity to determine the degree of matching between the script and the topic. The pacing rationality index is based on the degree of fit between historical live broadcast pacing data and the audience attention curve.

[0010] Preferably, the reinforcement learning algorithm of the dynamic optimization unit performs the following steps: Receive audience sentiment analysis results from the real-time feedback system; If the proportion of negative emotions exceeds the threshold, an optimization instruction is triggered; Adjust the product display order, interactive density, or promotional message intensity in the script according to the optimization instructions.

[0011] Preferably, the multidimensional analysis module includes: The audience profiling analysis submodule constructs a user segmentation model based on the K-means clustering algorithm; The Creative Popularity Prediction Submodule uses the ARIMA time series model to predict the popularity trend of creative keywords. The business value assessment submodule establishes a linear regression model between product conversion rate and script elements.

[0012] A method for intelligent optimization of live streaming scripts includes the following steps: S1: Collect audience data, live stream script data, and historical feedback data; S2: A script generation framework based on LSTM neural network; S3: Quantitatively score the script using a multi-indicator fusion model; S4: Receive audience feedback during the live stream and dynamically adjust the script content and pace based on reinforcement learning algorithms; S5: Outputs the optimized script and a business value assessment report.

[0013] Preferably, the calculation formula for the multi-index fusion model in step S3 is as follows: Q = w1 × G + w2 × R + w3 × P; Where Q is the script quality score, G is the grammatical accuracy, R is the content relevance, P is the pacing rationality, and w1, w2, and w3 are weighting coefficients.

[0014] Preferably, the dynamic adjustment in step S4 includes: When negative emotions from the audience account for more than 30%, insert limited-time offer messages into the script; When the rate of decrease in audience dwell time exceeds 20%, the frequency of interactive sessions should be increased. When a product's conversion rate is lower than its historical average, the product display order will be adjusted.

[0015] The present invention can achieve the following effects: Efficiency Improvement: Script generation efficiency increased by over 80%, and manual review workload reduced by 70%; Quality optimization: Script content relevance improved by 40%, and viewer dwell time increased by 25%; Business value-added: E-commerce live streaming conversion rate increases by 15%-30%; Standardized output: Establish a unified script quality assessment system for the industry. Attached Figure Description

[0016] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a data flow diagram of the present invention; Figure 3 This is a diagram of the algorithm model of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0018] Example 1: A live streaming script intelligent optimization system, the architecture of which includes: The multi-source data acquisition module is used to collect audience data, live stream script data, and historical feedback data; among them, Viewer data collected includes region, age, gender, viewing duration, number of interactions, and spending amount; live stream script data includes theme, duration, and number of products; historical feedback data includes feedback type, feedback content, feedback time, processing status, satisfaction level, and reasons for annotation. The script optimization engine includes: a content generation unit that generates a script framework based on the live stream topic, product information, and target audience characteristics using an LSTM neural network; the application of the LSTM network in content generation; data preprocessing: converting text data such as the live stream topic, product information, and target audience characteristics into word vector representations, for example, using Word2Vec or GloVe models; model training: training the LSTM network using labeled live stream script data to learn the grammatical, semantic, and structural features in the text sequence; script generation: based on the input live stream topic, product information, and target audience characteristics, the LSTM network outputs the next most likely word or phrase, gradually building a complete script framework.

[0019] The quality assessment unit uses a multi-indicator fusion model to quantitatively score the scripts; the multi-indicator fusion model includes a legal accuracy model, a content relevance model, and a pacing rationality model to quantitatively score the scripts.

[0020] The dynamic optimization unit adjusts the script content and pacing based on reinforcement learning algorithms and real-time audience feedback; real-time audience feedback includes sentiment analysis of bullet comments and dwell time, etc. The multi-dimensional analysis module is used for audience profiling, creative popularity prediction, and commercial value assessment. Audience profiling analysis: Based on clustering algorithms, user segmentation models are constructed, and personalized script suggestions are output; Creative popularity prediction: Predict the popularity trend of creative keywords through time series models, such as the ARIMA model; Business value assessment: Establish a regression model between product conversion rate and script elements, where the script elements are promotional message density and interaction frequency.

[0021] The real-time feedback system receives audience feedback during the live stream and drives the script optimization engine to make dynamic adjustments. Audience feedback includes bullet comments, likes, and purchasing behavior. The system drives the script optimization engine to adjust the live stream segments in real time, and makes dynamic adjustments such as increasing the frequency of interaction and adjusting the order of product display.

[0022] Furthermore, the multi-source data acquisition module includes: The audience data collection submodule collects audience data such as region, age, gender, viewing duration, number of interactions, and spending amount. The script data collection submodule collects information such as the live stream topic, script duration, number of products, and estimated sales revenue. The feedback data collection submodule collects viewer comments, likes, purchase behavior, and satisfaction ratings.

[0023] Furthermore, the multi-index fusion model of the quality assessment unit includes: Syntax accuracy metrics, based on NLP syntax tree analysis; Content relevance metrics are calculated using cosine similarity to determine the degree of matching between the script and the topic. The pacing rationality index is based on the degree of fit between historical live broadcast pacing data and the audience attention curve.

[0024] Furthermore, the reinforcement learning algorithm of the dynamic optimization unit performs the following steps: Receive audience sentiment analysis results from the real-time feedback system; If the proportion of negative emotions exceeds the threshold, an optimization instruction is triggered; Adjust the product display order, interactive density, or promotional message intensity in the script according to the optimization instructions.

[0025] Furthermore, the multidimensional analysis module includes: The audience profiling analysis submodule constructs a user segmentation model based on the K-means clustering algorithm; The Creative Popularity Prediction Submodule uses the ARIMA time series model to predict the popularity trend of creative keywords. The business value assessment submodule establishes a linear regression model between product conversion rate and script elements.

[0026] Example 2: A method for intelligent optimization of live streaming scripts, comprising the following steps: S1: Collect audience data, live stream script data, and historical feedback data; S2: A script generation framework based on LSTM neural network; S3: Quantitatively score the script using a multi-indicator fusion model; S4: Receive audience feedback during the live stream and dynamically adjust the script content and pace based on reinforcement learning algorithms; S5: Outputs the optimized script and a business value assessment report.

[0027] The calculation formula for the multi-index fusion model in step S3 is as follows: Q = w1 × G + w2 × R + w3 × P; Furthermore, Q represents the script quality score, G represents grammatical accuracy, R represents content relevance, P represents pacing rationality, and w1, w2, and w3 represent weighting coefficients.

[0028] Furthermore, the dynamic adjustment described in step S4 includes: When negative emotions from the audience account for more than 30%, insert limited-time offer messages into the script; When the rate of decrease in audience dwell time exceeds 20%, increase the frequency of interactive sessions; When a product's conversion rate is lower than its historical average, adjust the product display order.

[0029] Specific Implementation: E-commerce Live Streaming Script Optimization Process 1. Data Input: Users input the live stream topic (e.g., "New Beauty Product Launch"), product list (lipstick, foundation, etc.), and target audience (women aged 18-25); 2. Script generation: Content generation unit output script framework: 00:00-00:05 Opening greetings (quoting the popular topic "8 AM makeup") 00:05-00:15 Product 1 Display (Lipstick, highlighting the "long-lasting and non-fading" selling point) 00:15-00:25 Interactive segment (live stream lottery, keyword "lipstick swatches") Quality assessment unit scores: grammatical accuracy 98%, content relevance 92%, pacing appropriateness 85%.

[0030] 3. Dynamic optimization: Negative sentiments in viewer comments were detected during the live stream (the percentage of comments saying "the price is too high" increased). The real-time feedback system triggers an optimization instruction: insert a limited-time offer message ("First 100 orders receive a free makeup brush") into the script. Adjust the product display order to prioritize the display of high-value products.

[0031] 4. Output Effects: After optimization, the conversion rate increased by 22% and the audience dwell time increased by 18%.

[0032] Rhythm control algorithm implementation: Input: Historical live stream timing data (start / end time, activity type, popularity index); predict audience attention curve based on LSTM; dynamically insert high-popularity segments (such as raffles, Q&A); Output: Generates an optimized script with 6 rhythm points, increasing audience interaction frequency by 35%; Reinforcement learning algorithms: Reinforcement learning systems typically include the following five key elements: Intelligent Agent The executor of decisions, such as the "dynamic optimization unit" in a live script optimization system.

[0033] Choose an action based on the current state (such as adjusting script content or inserting interactive elements).

[0034] Environment External systems for intelligent agent interaction, such as live streaming scenarios (including real-time data such as audience feedback and product conversion rates).

[0035] The environment will provide new status and reward signals based on the agent's actions.

[0036] State s ) The current description of the environment, such as: audience sentiment analysis results (positive / negative bullet comments ratio); real-time viewership, dwell time, product click-through rate, and other indicators.

[0037] Action a ) The intelligent agent can perform actions such as: inserting limited-time offer messages into the script; adjusting the order of product display; and increasing the frequency of interactive elements.

[0038] Reward r ) The environment provides feedback signals to the agent's actions, such as: positive rewards: increased viewer dwell time and improved conversion rate; negative rewards: increased proportion of negative comments and increased viewer churn rate.

[0039] The interactive process of reinforcement learning can be summarized as the following loop: Observation state: The current state of the environment perceived by the agent. st (e.g., audience sentiment analysis results).

[0040] Choose an action: based on the policy. π Select Action at (e.g., inserting interactive elements).

[0041] Execution action: Environment reception action at , transition to a new state st +1, and feedback reward rt .

[0042] Update strategy: The agent updates its strategy based on rewards. rt and new status st +1 Optimization Strategy π To maximize long-term rewards.

[0043] Mathematical representation: The goal is to maximize the expected value of cumulative rewards: in ∈[0,1] is a discount factor that balances immediate rewards with long-term rewards.

[0044] The implementation steps of the reinforcement learning algorithm in the live streaming script intelligent optimization system are as follows: State definition Input features include: Audience sentiment analysis results (such as the percentage of negative comments); Real-time viewing data (duration of stay, interaction rate); Product conversion rate, click-through rate, and other business metrics.

[0045] Action space design Executable actions: Adjust the script content (e.g., insert promotional phrases); Modify the pacing (e.g., add interactive elements); Optimize the product display order.

[0046] Reward function design Positive Rewards: The duration of visitor stay increased (+1 minute / minute); Product conversion rate improved (+5 points / 1% growth).

[0047] Negative Rewards: The proportion of negative comments increased (-2 points / 1% increase); The audience dropout rate increased (-3 points / 1% growth).

[0048] System architecture diagram: Front-end interaction layer: For users and operations personnel, it provides functions such as script uploading, optimization configuration, and result viewing; Business logic layer: core logic processing module, including script optimization engine, multi-dimensional analysis module and real-time feedback system; Data service layer: Provides access interfaces for audience, script, and feedback data, which can be called by the business logic layer; Infrastructure layer: includes data storage (such as databases, object storage), model training platforms (such as machine learning platforms), and real-time computing engines.

[0049] Data flow diagram: 1. Data Input: Audience behavior data (such as region, gender, viewing time, number of interactions, etc.) Script data (such as theme, duration, structure, etc.) Feedback data (such as bullet comments, likes, and purchasing behavior). 2. Data processing flow: Data acquisition module: Unifies the collection of data from multiple sources; Data cleaning and preprocessing: noise removal, handling missing values, and format standardization; Feature engineering: Extracting key features, such as user profiles, popularity metrics, and commercial value metrics; 3. Model Reasoning and Rule Engine: Based on feature input, the algorithm model or rule engine is invoked to make script optimization decisions.

[0050] 4. Real-time feedback and output: The real-time feedback system returns optimization instructions to the inference engine, forming a closed loop; The final output includes an optimized script and a business value assessment report.

[0051] Algorithm model diagram: Input layer Audience feature vectors include user profiles, interest tags, and behavioral sequences; Script feature vector: includes topic, duration, keywords, structure, etc.; Historical feedback characteristics: such as the sentiment of bullet comments, interaction rate, conversion rate, etc.

[0052] Model layer Audience profiling model: Using clustering or classification algorithms to construct user profiles; Creative popularity prediction model: predicts content popularity based on time series or deep learning models (such as LSTM, Transformer); Business valuation models: Evaluate the business potential of a script using regression or tree models (such as XGBoost); Reinforcement learning optimization strategy: The optimization process is modeled as an MDP (Markov Decision Process), and optimization scripts are generated through a policy network.

[0053] Output layer Script optimization: Generate optimal script content; Scoring and Confidence: Output script quality score and confidence level; Optimization suggestions: Provide suggestions for script improvement (such as adjusting the rhythm, enhancing interaction, etc.); Summarize System architecture diagram: shows the system's layered structure and module dependencies; Data flow diagram: Describes the complete flow of data from acquisition to output; Algorithm model diagram: This diagram illustrates the input, processing, and output structure of the core algorithm model.

[0054] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0055] In summary, the above description is only a specific embodiment of the present invention, but the structural features of the present invention are not limited thereto. Any changes or modifications made by those skilled in the art within the scope of the present invention are covered by the patent scope of the present invention.

Claims

1. A live streaming script intelligent optimization system, characterized in that, include: The multi-source data acquisition module is used to collect audience data, live stream script data, and historical feedback data. Script optimization engine, including: The content generation unit generates a script framework based on the LSTM neural network according to the live broadcast topic, product information, and target audience characteristics. The quality assessment unit uses a multi-indicator fusion model to quantitatively score the scripts; The dynamic optimization unit adjusts the script content and pacing based on reinforcement learning algorithms and real-time audience feedback. The multi-dimensional analysis module is used for audience profiling, creative popularity prediction, and commercial value assessment. The real-time feedback system receives audience feedback during the live stream and drives the script optimization engine to make dynamic adjustments.

2. The intelligent optimization system for live streaming scripts according to claim 1, characterized in that, The multi-source data acquisition module includes: The audience data collection submodule collects audience data such as region, age, gender, viewing duration, number of interactions, and spending amount. The script data collection submodule collects information such as the live stream topic, script duration, number of products, and estimated sales revenue. The feedback data collection submodule collects viewer comments, likes, purchase behavior, and satisfaction ratings.

3. The intelligent optimization system for live streaming scripts according to claim 1, characterized in that, The multi-indicator fusion model of the quality assessment unit includes: Syntax accuracy metrics, based on NLP syntax tree analysis; Content relevance metrics are calculated using cosine similarity to determine the degree of matching between the script and the topic. The pacing rationality index is based on the degree of fit between historical live broadcast pacing data and the audience attention curve.

4. The intelligent optimization system for live streaming scripts according to claim 1, characterized in that, The reinforcement learning algorithm of the dynamic optimization unit performs the following steps: Receive audience sentiment analysis results from the real-time feedback system; if the proportion of negative sentiment exceeds the threshold, trigger an optimization instruction; adjust the product display order, interactive density, or promotional intensity in the script according to the optimization instruction.

5. The intelligent optimization system for live streaming scripts according to claim 1, characterized in that, The multidimensional analysis module includes: The audience profiling analysis submodule constructs a user segmentation model based on the K-means clustering algorithm; the creative popularity prediction submodule predicts the trend of creative keyword popularity through the ARIMA time series model; and the business value assessment submodule establishes a linear regression model between product conversion rate and script elements.

6. A method for intelligent optimization of live streaming scripts, characterized in that, Includes the following steps: S1: Collect audience data, live stream script data, and historical feedback data; S2: A script generation framework based on LSTM neural network; S3: Quantitatively score the script using a multi-indicator fusion model; S4: Receive audience feedback during the live stream and dynamically adjust the script content and pace based on reinforcement learning algorithms; S5: Outputs the optimized script and a business value assessment report.

7. The method for intelligent optimization of live streaming scripts according to claim 6, characterized in that, The calculation formula for the multi-index fusion model in step S3 is as follows: Q = w1 × G + w2 × R + w3 × P; Where Q is the script quality score, G is the grammatical accuracy, R is the content relevance, P is the pacing rationality, and w1, w2, and w3 are weighting coefficients.

8. The method for intelligent optimization of live streaming scripts according to claim 6, characterized in that, The dynamic adjustment mentioned in step S4 includes: When negative emotions from viewers account for more than 30%, insert limited-time offer messages into the script; when the viewers' dwell time decreases by more than 20%, increase the frequency of interactive segments; when the product conversion rate is lower than the historical average, adjust the product display order.

Images