Multi-modal cross attention sentiment analysis of textual and audio embeddings
A method for managing sentiment analysis includes obtaining, by a data classification system, raw data, associated with a meeting, wherein the raw data the comprises a textual transcript and an audio file of the meeting, applying a generative adversarial network (GAN) to the audio file to obtain audio embeddings and extracted text, applying the textual transcripts and the extracted text to a trained language model to obtain textual embeddings, applying the audio embeddings and the textual embeddings to a multi-modal cross attention module to obtain a fused embedding, performing a sentiment classification on the fused embedding to obtain a sentiment prediction, and implementing a remediation on the data classification system using the sentiment prediction.
1 . A method for managing sentiment analysis, the method comprising:
obtaining, by a data classification system, raw data, associated with a meeting, wherein the raw data the comprises a textual transcript and an audio file of the meeting;
applying a generative adversarial network (GAN) to the audio file to obtain audio embeddings and extracted text;
applying the textual transcripts and the extracted text to a trained language model to obtain textual embeddings;
applying the audio embeddings and the textual embeddings to a multi-modal cross attention module to obtain a fused embedding;
performing a sentiment classification on the fused embedding to obtain a sentiment prediction; and
implementing a remediation on the data classification system using the sentiment prediction.
2 . The method of claim 1 , wherein the GAN comprises a sample generator and a sample discriminator.
3 . The method of claim 2 , wherein applying the GAN to the audio file comprises:
using data distribution of the audio file to obtain a set of audio samples;
generating, by the sample generator and using the data distribution, a set of synthetic samples;
applying the set of synthetic samples and the set of audio samples to the sample discriminator and modifying the set of synthetic samples in an iterative fashion until a pre-defined percentage threshold of the set of synthetic samples are labeled as not synthetic; and
obtaining the audio embeddings as a final modified set of synthetic samples.
4 . The method of claim 2 , wherein the sample generator has a higher learning rate than that of the sample discriminator.
5 . The method of claim 1 , wherein the sentiment prediction comprises tagged textual embeddings of the textual transcript, wherein the tagged textual embeddings comprises at least one of: a positive tagged embedding, a negative tagged embedding, a very positive tagged embedding, a very negative tagged embedding, and a neutral tagged embedding.
6 . The method of claim 1 , wherein the trained language model is an auto-regressive language model trained using a set of earnings call transcripts.
7 . The method of claim 1 , wherein one of the fused embeddings is tagged with a pitch, frequency, and tone associated with a textual embedding based on a corresponding audio embedding.
8 . A non-transitory computer readable medium comprising computer readable program code, which when executed by a computer processor enables the computer processor to perform a method for managing sentiment analysis, the method comprising:
obtaining, by a data classification system, raw data, associated with a meeting, wherein the raw data the comprises a textual transcript and an audio file of the meeting;
applying a generative adversarial network (GAN) to the audio file to obtain audio embeddings and extracted text;
applying the textual transcripts and the extracted text to a trained language model to obtain textual embeddings;
applying the audio embeddings and the textual embeddings to a multi-modal cross attention module to obtain a fused embedding;
performing a sentiment classification on the fused embedding to obtain a sentiment prediction; and
implementing a remediation on the data classification system using the sentiment prediction.
9 . The non-transitory computer readable medium of claim 8 , wherein the GAN comprises a sample generator and a sample discriminator.
10 . The non-transitory computer readable medium of claim 9 , wherein applying the GAN to the audio file comprises:
using data distribution of the audio file to obtain a set of audio samples;
generating, by the sample generator and using the data distribution, a set of synthetic samples;
applying the set of synthetic samples and the set of audio samples to the sample discriminator and modifying the set of synthetic samples in an iterative fashion until a pre-defined percentage threshold of the set of synthetic samples are labeled as not synthetic; and
obtaining the audio embeddings as a final modified set of synthetic samples.
11 . The non-transitory computer readable medium of claim 9 , wherein the sample generator has a higher learning rate than that of the sample discriminator.
12 . The non-transitory computer readable medium of claim 8 , wherein the sentiment prediction comprises tagged textual embeddings of the textual transcript, wherein the tagged textual embeddings comprises at least one of: a positive tagged embedding, a negative tagged embedding, a very positive tagged embedding, a very negative tagged embedding, and a neutral tagged embedding.
13 . The non-transitory computer readable medium of claim 8 , wherein the trained language model is an auto-regressive language model trained using a set of earnings call transcripts.
14 . The non-transitory computer readable medium of claim 8 , wherein one of the fused embeddings is tagged with a pitch, frequency, and tone associated with a textual embedding based on a corresponding audio embedding.
15 . A system, comprising:
a processor; and
memory including instructions, which when executed by the processor, perform a method comprising:
obtaining, by a data classification system, raw data, associated with a meeting, wherein the raw data the comprises a textual transcript and an audio file of the meeting;
applying a generative adversarial network (GAN) to the audio file to obtain audio embeddings and extracted text;
applying the textual transcripts and the extracted text to a trained language model to obtain textual embeddings;
applying the audio embeddings and the textual embeddings to a multi-modal cross attention module to obtain a fused embedding;
performing a sentiment classification on the fused embedding to obtain a sentiment prediction; and
implementing a remediation on the data classification system using the sentiment prediction.
16 . The system of claim 15 , wherein the GAN comprises a sample generator and a sample discriminator, and wherein the sample generator has a higher learning rate than that of the sample discriminator.
17 . The system of claim 16 , wherein applying the GAN to the audio file comprises:
using data distribution of the audio file to obtain a set of audio samples;
generating, by the sample generator and using the data distribution, a set of synthetic samples;
applying the set of synthetic samples and the set of audio samples to the sample discriminator and modifying the set of synthetic samples in an iterative fashion until a pre-defined percentage threshold of the set of synthetic samples are labeled as not synthetic; and
obtaining the audio embeddings as a final modified set of synthetic samples.
18 . The system of claim 15 , wherein the sentiment prediction comprises tagged textual embeddings of the textual transcript, wherein the tagged textual embeddings comprises at least one of: a positive tagged embedding, a negative tagged embedding, a very positive tagged embedding, a very negative tagged embedding, and a neutral tagged embedding.
19 . The system of claim 15 , wherein the trained language model is an auto-regressive language model trained using a set of earnings call transcripts.
20 . The system of claim 15 , wherein one of the fused embeddings is tagged with a pitch, frequency, and tone associated with a textual embedding based on a corresponding audio embedding.