A 5 fold Question Answering DL based model to retrieve information from a data-set created for Farmers. Along with that a Express based back-end which answers queries from Google Actions Platform
Over 50% of the Indian population is engaged in agriculture. Coupled with
the statistic that casual workers in the agriculture sector have the lowest
level of digital literacy at 13%, it is vitally important that the agricultural
workforce have access to trustable, accurate information on farming. One
of the most easily accessible means of device communication is via voice
and voice based Assistants like Siri, Google Assistant and Amazon Alexa.
While multiple attempts have been made to try and incorporate the two - a
voice based information system with the farming community, there still
remain unanswered questions particularly around the HCI aspect of the
design and moreover, around the utility and actionability of the
information provided by those systems. This project aims to extend upon
and re-establish upon those questions.
To build a Voice-based Information system to
disseminate agricultural information to farmers.
• To further build upon a the audio-visual UI and
to make it more trustworthy, friendly, easy to
use based on the metrics from the Literature
Review
• To create a product that aims to focus on three
primary metrics of success
– Task Completion Rate - this metric will
measure the percentage of successful usages that our target audience has with the
chatbot
– System Usability Scale - measuring ease of
use via the NASA-TLX Scale
We already have a pre-existing model of interface tested on 5 sources based on a preliminary model of trust. The pre-existing model focused on the visual representation of the interface and how our target group would interact with the model as a whole. Hence, we currently have a chatbot that is appealing on it’s first interaction with the target audience. Now, we have to work on making it appealing over a longer period of time as well as making it easier to use. We aim to accomplish this by testing the granular interactions between the target audience and the chatbot. Last semester, the chatbot was optimized from a macro perspective. Our plan this semester was to test individual interactions and judge them primarily on the metrics given below.
To focus on developing the pre-existing environment over the following 7 parameters:
User-testing on response to different voices
User-testing on response to different avatar responses
User-testing and deployment on different phrasing mechanisms focusing on three subparameters
• User-testing to evaluate adoption mechanisms on the following parameter
• User-testing on integrating an entertainment element into initial chatbot interaction
User-testing over response to integrating with pre-existing farmer resources
User-testing across the response to varied dialects and focusing on how regionality affects the openness
There are 3 primary factors that any future study on this topic or the continuation of this project will have to contend with. • Future groups should attempt to create the Information System in isolation without attempting to provide a cross-adoption vertical like we did with entertainment. The farmer groups who answered the entertainment survey told us that farmers did not need additional outlets to use information apps as their usages of information apps and entertainment apps are quite disjoint in nature. They told us that even in their past attempts at using agricultural based voice information systems, they used it on it’s own merit and stopped when using the Action became too tricky for them. They did not need an Entertainment feature to keep them into using the chatbot. • Future groups pursuing this survey should attempt to link up the front-end and the backend prior to the conduction of the survey. This will aid in reducing technical issues during the survey while also providing a much clearer perspective over how the back-end models should be modified to better attune it to the requirements of the farmer groups. • While we are measuring for the influence of societal constructs like peer pressure, fear of power structures and likelihood of social alienation - we are not acting upon them where penetration strategies are concerned. Future groups would need to use that data to find how best to assuage the concerns of the user group and get them to use the chatbot without fear and apprehension over rebuke from older sources of information. This was one of the major issues that was faced in the second batch of survey collections and can easily taint the responses of the same.
Contact errijuldahiya(at)gmail.com to get more information on the project