Multi-Platform Sentiment Analysis Pipeline
Budget: ₹100 – ₹400 INR
I need a robust, ML-driven pipeline that continuously gathers conversations from Twitter, Facebook, Instagram, and the wider public web, then interprets them in two complementary ways: live streaming for immediate insights and batch processing for deep historical trends.
The core outputs I’m after are overall sentiment scores (positive, neutral, negative), fine-grained emotion tags such as happiness, anger, and sadness, plus an automatically refreshed list of trending keywords or topics. Accuracy, scalability, and low-latency delivery matter more to me than flashy visuals, yet I’d still like concise dashboards or API endpoints that make the results easy for downstream teams to consume.
Typical tooling for this kind of work might involve Python, spaCy or NLTK for text cleansing, a transformer model fine-tuned in TensorFlow or PyTorch, and a Kafka or WebSocket layer to keep the real-time feed flowing; feel free to propose equivalent technologies if you have a stronger approach. All code must be well-documented and deployable to a cloud environment such as AWS or GCP so I can spin up or scale down processing nodes as needed.
Acceptance criteria
• Real-time stream ingested and processed with end-to-end latency under one minute.
• Historical backfill capable of digesting at least twelve months of data per platform.
• JSON or CSV exports that include original text, platform source, timestamp, sentiment score, emotion label, and top keyword tags.
• Clear README and deployment scripts (Docker or Terraform preferred) so I can reproduce the environment from scratch.
If parts of the stack—data collection, model training, or visualisation—are your specialty, mention that; I’m open to phased delivery as long as each milestone is functional.
The core outputs I’m after are overall sentiment scores (positive, neutral, negative), fine-grained emotion tags such as happiness, anger, and sadness, plus an automatically refreshed list of trending keywords or topics. Accuracy, scalability, and low-latency delivery matter more to me than flashy visuals, yet I’d still like concise dashboards or API endpoints that make the results easy for downstream teams to consume.
Typical tooling for this kind of work might involve Python, spaCy or NLTK for text cleansing, a transformer model fine-tuned in TensorFlow or PyTorch, and a Kafka or WebSocket layer to keep the real-time feed flowing; feel free to propose equivalent technologies if you have a stronger approach. All code must be well-documented and deployable to a cloud environment such as AWS or GCP so I can spin up or scale down processing nodes as needed.
Acceptance criteria
• Real-time stream ingested and processed with end-to-end latency under one minute.
• Historical backfill capable of digesting at least twelve months of data per platform.
• JSON or CSV exports that include original text, platform source, timestamp, sentiment score, emotion label, and top keyword tags.
• Clear README and deployment scripts (Docker or Terraform preferred) so I can reproduce the environment from scratch.
If parts of the stack—data collection, model training, or visualisation—are your specialty, mention that; I’m open to phased delivery as long as each milestone is functional.