Team Name

Ping Pong

Team Members

Project Description

https://github.com/jessearcana/govhack2023/blob/main/README.md

Problem Statement

"What makes a great holiday?" is personal. Everyone has their own blend of unwinding and exploring. In South Australia, we understand that not everyone wants the same experience. Maybe the standard tourist packages don't work for you and you want to mix and match your own holiday. We also understand that not everyone has time to study travel guides and reviews -- a holiday is supposed to be a break from getting organised and not make more work for you.

South Australia has an enormous variety of holiday activities for you to enjoy. So many, in fact, that we think many people will be overwhelmed with choice.

Our Proposal

Introducing "Heaps Good". Heaps Good is a machine learning assistant that you can talk to to describe the sort of holiday activities that you want. Currently in proof-of-concept form, Heaps Good has a small database of travel activities, locations, and tourism providers that it uses to make recommendations based on what you request. It uses the latest technology in Natural Language Processing, vector databases, and similarity searching to tailor its recommendations to your requirements.

Description

Heaps Good analyses free text typed by the user. It uses large language models to create a response, and semantic embedding coupled with vector databases to ensure that the presented content is grounded to reality (LLMs have a habit of hallucinating imaginary answers). There are several benefits of our approach compared to current product offerings

benefit 1

Reuslts are customised to the user, and reflect inferred interests that the user might have.

benefit 2

Tourist research has become difficult because of the commercial capture of information sources. This has lead to tourists having a sub-optimal experience, as well as a tendency to funnel tourists into a few large, well-known, activies.
Tourism SA, because of its non-commercial nature, has an opportunity to build a portal that is designed with the tourist and the delivery of an excellent holiday as the principal design considerations. Such a portal is currently lacking because commercial systems tend to prioritise paying for product placement above the tourist experience.

benefit 3

Our portal will, over time, create a dataset of its own from which it can learn and thus improve performance

Commercial vision

We believe that Heaps Good can become the main way of planning holidays in South Australia, and a significant commercial advantage for the state. By keeping the portal itself non-commercial, it can focus on delivering the best outcome for tourists, leading to better experience, repeat custom, and word-of-mouth effects.

Envisioned Improvements

This is currently a proof-of-concept product only.

There are many possibilities to extend this platform to improve usability. Some ideas are that it more-effectively considers:
- transport between activities, inlcuding mapping
- client classification to make inferences about wants (for example, clients who play golf might also like drinking wine, whereas clients who enjoy go-kart racing might prefer beer). There are thousands of similar inferences that can be made by analysing the communication received from the client - the words they use, the manner of expression, the sound of their voice, their travel companions, etc
- finding "Hidden Gems" (for those who want them), such as the rock pool at Innes National Park, Stokes Bay, or Skillogalee Winery
- engage in discourse rather than a single request. The client will be able to say "I want more of that" or "I won't want to do that because", "I want something more cultural", "I want somehting random", and the system will adapt its proposal accordingly
- making a learning system that understands the sorts of recommendations that different people respond well to
We have the knowledge and skill to implement all of the above, but have not done so due to time constraints.
- improving the database of available destinations

Featured datasets

last slide

Heaps Good has been made by the Adelaide consultancy, Apex Arcana. We specialise in data science and helping people to make the most of their data. Please talk to use about how we can help you.

Data Story

data from the google places API, and simulated data which could be provided from the ATDW (which is not currently open), is indexed in a vector index. User provided holiday descriptions are converted into a story by Open AI then key activities and places are matched with the vector data to retain the compelling story telling but bring it back to real data that can be relied upon (and curated if need be).