ABOUT BAZAAR

Bazaar Graphing is part of an exploratory series on reimagining ecommerce patterns and experiences. Search results are represented in a network graph and clicking on a product will generate more semantically similar products.

This series (and the name) is inspired in part by Eric S. Raymond's seminal essay 'The Cathedral and the Bazaar'and in part by consistently finding myself 75 pages deep in SSENSE sale twice a year.

BROWSE AND SEARCH

The current paradigm in which we search for product online is enabled through keyword search. Products, via their titles, descriptions and meta data, tags…and I'm sure there are others, contain key words that businesses think customers will use to search for said product. This system assumes that the 'search' surface area is narrow only allowing your query to hit the exact keyword in order to surface product, creating a very narrow and exacting result list.

Very specific behaviours emerge in this way when we shop online and limits us to simply browse or search. Either we click through countless pages to find a product we like, or we enter a specific search into a search bar hoping for the right product. (Don't even get me started on the vagaries of Google searching products…)

SHIFTING FROM SEARCH TO DISCOVERY

With the advent of LLMs and the possibility of integrating augmented cognition into existing processes, there's an opportunity to increase the footprint of that specific touchpoint into something 'fuzzier' via semantic search. Not only does this surface the exact result but also similar results that fall within the range. This enables a squishier search deepening and personalizing experiences. What this enables is a new way to search, moving away from Search to Discovery and enabling surfacing of product that is not the same but similar.

GOING BEYOND CHAT PATTERNS AND GRID UIS

The UI design for this project aims to explore typologies and UI patterns that are distinctly non-chat interface. Current AI-app paradigms rely mostly on conversational exchanges but for the purpose of exploration, we head in the same direction of work from companies like WebSim, ArcBrowser and Globe Explorer who seek to Interact with AI in novel ways. The purpose in this case is to abstract away from the explicit idea of AI and all the baggage of hype and skepticism that comes with it, to focus on the experience.

GRAPH AS UI

The graphing UI mirrors the architecture of the system itself, product descriptions are embedded into a vector space where similar products are clustered together, each product becomes a node or point in this cartesian space. When users query the database via natural language, this request gets upserted and the most similar product nodes around it are then surfaced for the user.

CLICKING AS A METHOD OF DISCOVERY

Without a chat interface, we use 'clicking' on a product to trigger generation of semantically similar products, once clicked, the system will generate 3-5 semantically similar products from that node until there are no more results available. Since results are semantically similar, there might be products surfaced that are a surprise allowing you to explore in the direction that most interests you.

BREADCRUMBS FOR CONTEXTUAL TRAIL

Searching additional queries will add new nodes to your graphing space even if they are not semantically related allowing you to visually keep your search history and maintain contextual awareness, like breadcrumbs if you will.

xoxo
Sharon

← Return to Graph Explorer