Systems and methods for generating video suggestions
A method for generating video suggestions to users of an online video streaming platform is provided. A video suggestion engine may scan and analyze the comments section of an online video to generate a sentiment score associated with the video. The sentiment score may be indicative of the level of association between a given sentiment and a given video. After, the video suggestion engine may suggest the video to a user, based on the computed sentiment score. In some cases, more than one sentiment score may be computed for a given video. In some cases, multiple sentiment scores may be computed for multiple videos. In some cases sentiment scores across videos can be compared to identify a video for suggesting to a user, and the method may generate a video suggestion based on such a comparison.
1 . A method for suggesting a video for viewing to a user of a video streaming platform, the method comprising:
acquiring first comments from a first comments section associated with a first video;
applying a first artificial intelligence (AI) model to the first comments to identify one or more first sentiment correlation indicators indicating a correlation between the first video and a sentiment, wherein the first AI model is trained, using historical user-generated comments labeled by a second AI model, to produce sentiment correlation indicators for a given video with an associated comments section;
generating, using the one or more first sentiment correlation indicators, a first video sentiment score quantifying the correlation between the first video and the sentiment; and
providing, using the first video sentiment score, the first video as a suggested video for viewing to the user of the video streaming platform.
2 . The method of claim 1 , wherein a sentiment correlation indicator of the one or more first sentiment correlation indicators comprises a portion of a comment of the first comments that indicates a correlation between the first video and the sentiment.
3 . The method of claim 1 , wherein the first video sentiment score comprises a measure of an amount of comments within the first comments section associated with the first video that indicate a correlation between the first video and the sentiment.
4 . The method of claim 1 , wherein the sentiment comprises at least one of happiness, nostalgia, comedy, romance, or a sentiment associated with a genre of music.
5 . The method of claim 1 , further comprising:
acquiring second comments from a second comments section associated with a second video;
applying the first AI model to the second comments to identify one or more second sentiment correlation indicators indicating a correlation between the second video and the sentiment;
generating, using the one or more identified second sentiment correlation indicators, a second video sentiment score quantifying the correlation between the second video and the sentiment; and
providing, the first video as a suggested video for viewing to the user of the video streaming platform, responsive to a comparison between the first video sentiment score and the second video sentiment score.
6 . The method of claim 5 , wherein providing the first video as the suggested video is based at least on a determination that the first video sentiment score is higher than the second video sentiment score.
7 . The method of claim 1 , wherein the first AI model is pre-trained to identify sentiment correlation indicators within natural language text.
8 . The method of claim 1 , wherein the first comments comprise one or more user-posted emojis associated with the first comments section.
9 . The method of claim 8 , wherein the first AI model is trained by:
generating a first training dataset comprising:
a first plurality of training comments, and
a first plurality of ground truth labels indicating whether a given comment of the first plurality of training comments correlates with the sentiment, wherein at least a subset of the first plurality of ground truth labels is generated using the second AI model; and
training the first AI model using the first training dataset.
10 . The method of claim 9 , wherein the second AI model is trained by:
generating a second training dataset comprising:
a second plurality of training comments, and
a second plurality of human-annotated ground truth labels indicating whether a respective comment of the second plurality of training comments correlates with the sentiment; and
training the second AI model using the second training dataset.
11 . The method of claim 9 , wherein training the first AI model using the first training dataset comprises:
processing, using the first AI model, the given comment to generate a training label indicating whether the given comment correlates with the sentiment;
identifying a difference between the training label and the ground truth label of the first plurality of ground truth labels; and
changing one or more parameters of the first AI model to reduce or eliminate the identified difference.
12 . The method of claim 9 , wherein the first AI model is trained until the first AI model achieves an accuracy threshold on a validation dataset.
13 . The method of claim 9 , wherein a number of parameters of the second AI model is at least a predefined number of times larger than a number of parameters of the first AI model.
14 . A method for suggesting a video for viewing to a user of a video streaming platform, the method comprising:
acquiring video content data associated with a video;
applying a first artificial intelligence (AI) model to the video content data to generate a video sentiment score quantifying a correlation between the video and a sentiment, wherein the first AI model is trained, using historical video content data comprising user-generated comments labeled by a second AI model, to produce sentiment correlation indicators for a given video with associated video content data; and
providing, using the video sentiment score, the video as a suggested video for viewing to the user of the video streaming platform.
15 . The method of claim 14 , further comprising:
generating a first training dataset comprising:
a plurality of training comments, wherein each comment of the plurality of training comments is associated with a video of a plurality of videos, and
a first plurality of ground truth labels indicating whether a given comment of the plurality of training comments correlates a respective associated video of the plurality of videos with the sentiment, wherein at least a subset of the first plurality of ground truth labels is generated using the second AI model;
generating, using the first training dataset, a plurality of sentiment scores quantifying a correlation between a given video of the plurality of videos and the sentiment;
generating a second training dataset comprising:
a plurality of training video content data, and
a second plurality of ground truth labels indicating the sentiment score of the plurality of sentiment scores quantifying a correlation between a given video content data of the plurality of training video content data and the sentiment; and
training the first AI model using the second training dataset.
16 . The method of claim 15 , wherein training the first AI model using the second training dataset comprises:
processing, using the first AI model, the given video content data to generate a video sentiment score quantifying the correlation between the given video content data and the sentiment;
identifying a difference between the generated video sentiment score and a respective ground truth label of the second plurality of ground truth labels; and
changing one or more parameters of the first AI model to reduce or eliminate the identified difference.
17 . A system for suggesting a video for viewing to a user of a video streaming platform, the system comprising:
a memory device; and
a processing device communicatively coupled to the memory device, wherein the processing device is to:
acquire first comments from a first comments section associated with a first video;
apply a first artificial intelligence (AI) model to the first comments to identify one or more first sentiment correlation indicators indicating a correlation between the first video and a sentiment, wherein the first AI model is trained, using historical user-generated comments labeled by a second AI model, to produce sentiment correlation indicators for a given video with an associated comments section;
generate, using the one or more first sentiment correlation indicators, a first video sentiment score quantifying the correlation between the first video and the sentiment; and
provide, using the first video sentiment score, the first video as a suggested video for viewing to the user of the video streaming platform.
18 . The system of claim 17 , wherein the processing device is to further:
acquire second comments from a second comments section associated with a second video;
apply the first AI model to the second comments to identify one or more second sentiment correlation indicators indicating a correlation between the second video and the sentiment;
generate, using the one or more identified second sentiment correlation indicators, a second video sentiment score quantifying the correlation between the second video and the sentiment; and
provide, the first video as a suggested video for viewing to the user of the video streaming platform, responsive to a comparison between the first video sentiment score and the second video sentiment score.
19 . The system of claim 17 , wherein the first AI model is trained by:
generating a first training dataset comprising:
a first plurality of training comments, and
a first plurality of ground truth labels indicating whether a given comment of the first plurality of training comments correlates with the sentiment, wherein at least a subset of the first plurality of ground truth labels is generated using the second AI model; and
training the first AI model using the first training dataset.
20 . The system of claim 19 , wherein the second AI model is trained by:
generating a second training dataset comprising:
a second plurality of training comments, and
a second plurality of human-annotated ground truth labels indicating whether a respective comment of the second plurality of training comments correlates with the sentiment; and
training the second AI model using the second training dataset.