Closed-caption processing using machine learning for media advertisement detection
A method is provided to identify whether video content, which includes a plurality of image frames, is likely to include an advertisement. The video content is split into a plurality of segments, each segment having a pre-specified duration. Subtitle text information is extracted from each segment and is passed through a natural language processing (NLP) language model to extract an embedding representing the subtitle text information for each of the segments, wherein the NLP language model is previously trained to differentiate between subtitle text information from video content items that were each previously identified as being an advertisement in comparison to subtitle text information from video content items that were each previously identified as not being an advertisement. The embedding representing the subtitle text information for each of the segments is passed through a classifier to obtain a probability regarding whether each segment is an advertisement or not.
1. A computer-implemented method for building a model to classify whether video content is likely to be an advertisement, the video content including a plurality of image frames, the method comprising:
(a) providing a curated database of video content items that includes (i) a plurality of different video content items that were each previously identified as being an advertisement, and (ii) a plurality of different video content items that were each previously identified as not being an advertisement;
(b) splitting each of the video content items into a plurality of segments, each segment having a pre-specified duration;
(c) extracting subtitle text information from each segment;
(d) training a natural language processing (NLP) language model to differentiate between subtitle text information from video content items that were each previously identified as being an advertisement in comparison to subtitle text information from video content items that were each previously identified as not being an advertisement;
(e) passing the subtitle text information for each segment through the NLP language model to extract an embedding representing the subtitle text information for each of the segments; and
(f) passing the embedding representing the subtitle text information for each of the segments through a classifier to obtain a probability regarding whether each segment is an advertisement or not.
2. The method of claim 1 further comprising:
(g) aggregating the probabilities for each of the segments by a majority voting to classify whether the video content is likely to be an advertisement.
3. The method of claim 1 wherein the video content items that were previously identified as being an advertisement are split into segments of 15 seconds, and the video content items that were previously identified as not being an advertisement are split into segments of 10-15 seconds.
4. The method of claim 1 wherein the classifier is a softmax function that outputs a softmax probability.
5. A computer-implemented method for identifying whether video content is likely to include an advertisement, the video content including a plurality of image frames, the method comprising:
(a) splitting the video content into a plurality of segments, each segment having a pre-specified duration;
(b) extracting subtitle text information from each segment;
(c) passing the subtitle text information for each segment through a natural language processing (NLP) language model to extract an embedding representing the subtitle text information for each of the segments, wherein the NLP language model is previously trained to differentiate between subtitle text information from video content items that were each previously identified as being an advertisement in comparison to subtitle text information from video content items that were each previously identified as not being an advertisement; and
(d) passing the embedding representing the subtitle text information for each of the segments through a classifier to obtain a probability regarding whether each segment is an advertisement or not.
6. The method of claim 5 further comprising:
(e) aggregating the probabilities for each of the segments by a majority voting to classify whether the video content is likely to be an advertisement.
7. The method of claim 5 wherein the video content items that were previously identified as being an advertisement are split into segments of 15 seconds, and the video content items that were previously identified as not being an advertisement are split into segments of 10-15 seconds.
8. The method of claim 5 wherein the classifier is a softmax function that outputs a softmax probability.