{"id":28328699,"url":"https://github.com/akshaysinhaaa/emova","last_synced_at":"2026-05-05T23:31:10.765Z","repository":{"id":287028572,"uuid":"961525354","full_name":"akshaysinhaaa/emova","owner":"akshaysinhaaa","description":"A deep learning framework designed for emotion and sentiment recognition using text, audio, and video modalities. This project leverages the MELD (Multimodal EmotionLines Dataset) to train a robust and flexible model that reflects human communication more accurately than unimodal models.","archived":false,"fork":false,"pushed_at":"2025-05-24T19:32:32.000Z","size":32,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":1,"default_branch":"main","last_synced_at":"2025-05-24T20:31:42.915Z","etag":null,"topics":["bert","cnn","cuda","deep-learning","multimodal","python","pytorch","resnet-18","tensorboard","transformers"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/akshaysinhaaa.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2025-04-06T17:39:12.000Z","updated_at":"2025-05-24T19:32:36.000Z","dependencies_parsed_at":"2025-04-09T16:31:38.890Z","dependency_job_id":"20abb4a9-c449-483f-b0db-c02da5902c99","html_url":"https://github.com/akshaysinhaaa/emova","commit_stats":null,"previous_names":["akshaysinhaaa/sentiment-analysis","akshaysinhaaa/emova"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/akshaysinhaaa/emova","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/akshaysinhaaa%2Femova","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/akshaysinhaaa%2Femova/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/akshaysinhaaa%2Femova/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/akshaysinhaaa%2Femova/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/akshaysinhaaa","download_url":"https://codeload.github.com/akshaysinhaaa/emova/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/akshaysinhaaa%2Femova/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":32672499,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-05T11:29:49.557Z","status":"ssl_error","status_checked_at":"2026-05-05T11:29:48.587Z","response_time":54,"last_error":"SSL_connect returned=1 errno=0 peeraddr=140.82.121.5:443 state=error: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["bert","cnn","cuda","deep-learning","multimodal","python","pytorch","resnet-18","tensorboard","transformers"],"created_at":"2025-05-26T08:16:40.056Z","updated_at":"2026-05-05T23:31:10.754Z","avatar_url":"https://github.com/akshaysinhaaa.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# ConvEmoSentNet: A Parameter-Efficient Framework for Multimodal Emotion and Sentiment Analysis in Social Media Conversations\n\nA deep learning framework designed for **emotion and sentiment recognition** using **text**, **audio**, and **video** modalities. This project leverages the **MELD (Multimodal EmotionLines Dataset)** to train a robust and flexible model that reflects human communication more accurately than unimodal models.\n\n---\n\n## 📦 Dataset: MELD\n\n**Multimodal EmotionLines Dataset (MELD)** is a large-scale, multi-party conversation dataset derived from the TV series *Friends*. It provides aligned and synchronized **text**, **audio**, and **video** data, annotated with both **emotion** and **sentiment** labels.\n\n- **Modalities**:  \n  - `Text`: Dialogues (utterances)  \n  - `Audio`: Speaker voice tone  \n  - `Video`: Speaker facial expressions and posture\n\n- **Emotion Labels**:  \n  - Anger  \n  - Disgust  \n  - Fear  \n  - Joy  \n  - Neutral  \n  - Sadness  \n  - Surprise  \n\n- **Sentiment Labels**:  \n  - Positive  \n  - Negative  \n  - Neutral  \n\n🔗 [MELD Dataset GitHub](https://github.com/declare-lab/MELD)\n\n---\n\n## 🧠 Model Architecture\n\nThe model is **modular** and allows training on individual or fused modalities: `Text`, `Audio`, and `Video`. It is designed to perform well when one or more modalities are missing or unavailable.\n\n### 🔹 Individual Modality Encoders\n\n| Modality | Model Used         | Preprocessing                  |\n|----------|--------------------|--------------------------------|\n| Text     | BERT               | Tokenization, Padding          |\n| Audio    | CNN                | MFCC / Log-Mel Spectrogram     |\n| Video    | ResNet18 / 3D-CNN  | Face Extraction, Frame Sampling|\n\n### 🔹 Multimodal Fusion Strategy\n\n- Concatenation of latent vectors from each modality  \n- Optional **attention mechanism** to weight more informative modalities  \n- Final **Fully Connected Layers** leading to classification head (Softmax)\n\n```\n             ┌────────────┐    ┌────────────┐     ┌────────────┐\n             │   Text     │    │   Audio    │     │   Video    │\n             └────┬───────┘    └── ┬─────── ┘      └────┬───────┘\n                  │                │                   │\n                 BERT             CNN            3D CNN / ResNet\n                  │                │                   │\n                  └────────────┬───┴────┬──────────────┘\n                               │ Fusion │\n                               └────┬───┘\n                            Fully Connected\n                               Softmax\n```\n\n---\n\n## 🧪 Training Details\n\n- **Optimizer**: Adam\n- **Scheduler**: ReduceLROnPlateau  \n- **Loss Function**:  \n  - CrossEntropyLoss for multiclass emotion classification  \n  - Label Smoothing (0.1) to prevent overconfidence  \n- **Regularization**:  \n  - Dropout in FC layers (0.3–0.5)  \n  - Early Stopping based on validation loss  \n- **Batch Size**: 16–32  \n- **Epochs**: 15–25  \n\n### 🧵 Hyperparameter Tuning\n\n- Performed manually (grid search) on:\n  - Learning rate (1e-3 to 1e-5)  \n  - Hidden layer sizes  \n  - Dropout rates  \n  - Fusion strategies (early vs late fusion)\n\n---\n\n## 📈 Performance Snapshot\n\n| Configuration          | Emo Precision | Emo Acc. | Sen Precision | Sen Acc. |\n|------------------------|---------------|----------|---------------|----------|\n| Fused Model            | 53.50%        | 54.90%   | 64.40%        | 64.60%   |\n\n---\n\n## 🧑‍💻 Author\n\n**Akshay Sinha, Gauri Saksena, Yash Chandel**  \n_Deep Learning | Multimodal AI | Emotion Recognition_\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fakshaysinhaaa%2Femova","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fakshaysinhaaa%2Femova","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fakshaysinhaaa%2Femova/lists"}