App Store Review Analysis ✩
Personalised Assistance
Navigating websites and locating the right information can be challenging. Museum wanted to make it easier for the users to access personalized information for a better visitor experience. A LLM based solution was designed to address the issue and further evaluate with the end users for validation.
Before : User's POV
Overview
Improve visitor's access to information for better museum experience leading to greater engagement and satisfaction.
I built a llm based chatbot prototype to help the museum administration validate the use case so that the administration can create business case to implement llm based chatbot for their online customer enquiries and concierge.
I used RAG based architecture for LLM to process user query, search a vector database for relevant information and return responses in predefined formats using user's input language.
Review Categorization
Trigram Analysis
Before : User's POV
Overview
Improve visitor's access to information for better museum experience leading to greater engagement and satisfaction.
I built a llm based chatbot prototype to help the museum administration validate the use case so that the administration can create business case to implement llm based chatbot for their online customer enquiries and concierge.
I used RAG based architecture for LLM to process user query, search a vector database for relevant information and return responses in predefined formats using user's input language.
Ratings Over Time
Ratings Over Time
Key Components
Website Scraper
Website scraper to scrape data from existing website, generate embeddings and store it into a vector database.
Vector Search
Vector search to search across the database, retrieve results and rerank them as per relevance to the user query.
User Language
Detect the user language, search the specific database (e.g. EN for English and SC for Simplified Chinese) and summarize results in user's language.
Bot Templates
Formatting templates for bot to format its response specific to various use cases such as events, directions, follow-up queries or suggestions.
Website Scraper
Snippet of Python Code
Deployment Server
As per the usability test, all 24 users from Hong Kong and Mainland China used chatbot in different languages (English, Traditional Chinese and Simplified Chinese),and rated their experience with chatbot higher compared to their experience with the website.
This helped the museum administration to validate their hypotheses and build business case to get their project sponsored for LLM based chatbot.