Research

Possible solutions, what we chose, and why?: For our project, we had to choose from an array of different implementation options, as there exist three main types of chatbots, namely: rule-based, machine learning, and hybrid [4]. We settled for a task-based (rule-based) solution that utilizes natural language processing to extract the key information from a user's query and answer it appropriately. It does that by either recognizing the question as an FAQ or by utilizing its data retrieval and filtering capabilities. HeartBot is straightforward as it allows researchers and policy-makers to cherry-pick information from the BHF compendium, which saves them a lot of time and effort. We chose that solution because it is much easier to maintain than Machine learning alternatives, which require a bigger investment in both money and time [5]. Like all other software systems, HearBot has its limitations. That is why we decided to make it future-proof, by adding algorithms that allow the client to add more data whenever they need to, thus ensuring HeartBot can become 'smarter' on demand. There are also a number of auxiliary functions left in the source code to make the project extendable because we recognize that the company's needs may shift in the future, and other developers may decide to further our efforts.

Possible programming languages, frameworks, APIs, what we chose and why?: To complete the project we explored different libraries like AIML in Java, Chatterbot in Python, and web-based alternatives like Botpress and Botkit [3]. After extensive research, we realized that Chatterbot comes closest to what we want to achieve, and since Python was our mutual language of choice, we set out to write our chatbot code using Chatterbot with the idea to deploy a web app in Flask afterward. As the development process went on, however, we soon realized that Chatterbot wasn't a good match for our purposes. It was limited to logic adapters that weren't applicable to the data retrieval functionalities we needed to implement. Also, the machine learning engine required an enormous amount of dependencies and was running only on older versions of Python, which in turn meant that our project is going to become hard to maintain [2]. To fix that issue, we adapted our chatbot algorithm to work with a natural language processing toolkit from Python's nltk module. Nltk helped us develop the different components of HeartBot and compose a solution that first checks if the user question is an FAQ, and if not, creates a pandas query to extract the relevant data from the compendium in a filtered table. The query itself is constructed with the help of a python library called fuzzywuzzy, which utilizes the Levenshtein Distance to aid string matching [1]. The natural language processing toolkit and fuzzywuzzy were essential to properly categorizing user queries, and according to our research, were the only Python libraries that could meet all of the client's needs and allow room for future development. Lastly, we deployed our code to a web app using Flask. There is an alternative to Flask in Python, called Django, but Flask is much more lightweight and flexible. The learning curve for people who are new to web development is steeper in Django, so we settled on Flask because we knew we could utilize more functionalities with fewer lines of code [6].

[1] Cohen, A., 2022. fuzzywuzzy. [online] PyPI. Available at: [Accessed 8 March 2022].

[2] Cox, G., 2022. About ChatterBot — ChatterBot 1.0.8 documentation. [online] Chatterbot.readthedocs.io. Available at: [Accessed 8 March 2022].

[3] Patel, R., 2022. 10 Best AI Chatbot Development Frameworks [Pros + Cons + Integration +Pricing].[online]Space.Available At: [Accessed 8 March 2022].

[4] Team, F., 2022. 3 Types of chatbots you wish you knew sooner | Freshworks. [online]Freshworks.com.Available At: [Accessed 8 March 2022].

[5] Morozova, A., 2022. The cost of chatbot development — Jasoren. [online] Jasoren.com. Available at: [Accessed 8 March 2022].

[6] Singh, V., 2022. Flask vs Django in 2022: Which Framework to Choose?. [online] Hackr.io. Available at: [Accessed 8 March 2022].