Propaganda film content automatic generation method based on artificial intelligence

Abstract

A publicity film content automatic generation method based on artificial intelligence comprises the following steps: obtaining text data having a plurality of samples; data processing: eliminating sensitive words in the text data; generating a prompt term; dividing a data set, a training set and a test set; theme vectors: generating theme vectors for each sample, and embedding the theme vectors into a propaganda film generation model for training; and training a model by taking the prompt as model input and taking the propaganda film content as output, selecting a large model based on transformer for training, and generating the propaganda film content after the training is completed. According to the method, the propaganda film content can be conveniently and automatically generated, and high individuation is achieved.

Claims

1. A method for automatically generating promotional video content based on artificial intelligence, characterized in that: The following steps are involved: Get text data with several samples; Data processing, eliminating sensitive words in text data; Generate prompt words; Divide the data set into training set and test set; Topic vector: Generate a topic vector for each sample and embed it into the promotional video generation model for training; Use prompt as model input and promotional video content as output to train the model. Select a large transformer-based model for training. After training, generate promotional video content.

2. The method for automatically generating promotional video content based on artificial intelligence according to claim 1 is characterized in that: The data processing specifically includes the following steps: Data desensitization: use the spaCy library to regularize each sample, and then desensitize the private information, including age, name, address, and company name; Data cleaning, removing noise data, processing missing values ​​and outliers; Data integration: use the pandas library to integrate different data formats into a unified data format; Data annotation, marking the terms that reflect the characteristics of each sample.

3. The method for automatically generating promotional video content based on artificial intelligence according to claim 2 is characterized in that: When clearing data, use the smoothing method: remove the sudden change of noise and retain the trend information of the data by averaging, sliding average or weighted average of the data; Or the filtering method: The filtering method uses a filter to filter the data to remove the high-frequency components of the noise; Or it can be interpolation: interpolation inserts new data points between data points to fill in the missing values ​​caused by noise, making the data more continuous and smooth. Or outlier detection: outlier detection methods are used to identify and exclude outliers in data, which can be achieved through statistical analysis, outlier detection and outlier identification. Or it is a noise removal algorithm: The noise removal algorithm identifies and removes the effects of noise by calculating and analyzing the data.

4. The method for automatically generating promotional video content based on artificial intelligence according to claim 2 is characterized in that: The trained model outputs several promotional video templates, each of which is uniquely marked. When searching, by entering a keyword, the promotional video templates containing the keyword will pop up in a list for the user to view.

5. The method for automatically generating promotional video content based on artificial intelligence according to claim 2 is characterized in that: After processing the text data, the user types are obtained and a user type relationship diagram is generated; Perform multi-source graph fusion on the user type relationship graph, the preset user social relationship graph, the advertisement interaction graph, and the advertisement similarity graph to obtain a fusion result, and perform advertisement retrieval based on the fusion result to obtain an advertisement candidate set; The preset multi-task deep learning model is used to optimize the advertisement sorting and delivery of the advertisement candidate set to obtain the personalized promotional video content of the user.

6. The method for automatically generating promotional video content based on artificial intelligence according to claim 2 is characterized in that: When the topic vector is generated, it specifically includes: Use the tokenizer that comes with the BERT model to automatically segment the samples; Fill the length of the sample word vector to a fixed length and convert it to a tensor; Select the BERT-based large model as the topic vector model; Select hyperparameters such as batch size, learning rate, and number of training rounds; Select adamw as the optimizer for training; Train the topic vector model; After the model training is completed, the hidden state of the last layer is used as the topic vector.

7. The method for automatically generating promotional video content based on artificial intelligence according to claim 6 is characterized in that: When training a model, include: Use the tokenizer that comes with the big model to segment the text samples of the promotional video and convert the text data into indexes; Set the vocabulary size and maximum sequence length, pad or truncate the text to a fixed length (such as max_len), and convert to tensor; Use the trained topic vector model M1 to convert the effect into a topic vector; Combine the topic vector and the trailer text sequence as the input of the model; Select batch size, learning rate, and number of training rounds; Select adamw as the optimizer for training and the cross entropy loss function as the loss function; Use a learning rate scheduler or gradient clipping to stabilize the training process; After the model is trained, the similarity between the generated text and the reference text is evaluated using the BLEU and ROUGE indicators; Finally, generate the promotional video content.

8. The method for automatically generating promotional video content based on artificial intelligence according to claim 6 is characterized in that: After generating the promotional video content, the steps for automatic optimization include: Label each promotional video content, including product category, style, emotional tone, target audience and other dimensions; Analyze the results and infer user needs based on the scenario; Automatically adjust elements of the ad, including copy, visual design, layout, or presentation of the ad, to maximize fit and appeal with the user context.