As a search marketer, I came across the concept of commonsense reasoning in 2016 while reading articles on narrative intelligence. The curiosity to learn more about the topic and its key concepts led me to a thesis from Niket Tandon in 2019. His thesis is titled ‘Commonsense knowledge acquisition and applications.”
This initial exposure to commonsense reasoning and an increasing passion for graph theory led to my first presentation at BrightonSEO on the topic “From knowledge graphs to commonsense knowledge graphs.”
I continued to read more articles on commonsense reasoning and its importance in search. A simple Wikipedia entry on commonsense reasoning revealed the importance of how commonsense knowledge can be vital in understanding intentions. As a search marketer, it always leaves me with a degree of frustration when marketers and technologists reduce the user intent to the simplistic categories of informational. Navigational, transactional and commercial. This fails to reveal the ‘why’ behind the search and barely focuses on the “what.” It baffles me when established search marketing platforms add these simplistic intent labels next to keywords and almost overgeneralise the intent based on SERP results.
With further research on social psychology books like “Intention and intentionality: The Foundation of Social Cognition,” The Book of Why by Judea Pearl, Foundations of Statistics by Leonard Savage, Thinking Fast and Slow by Daniel Kahneman and a few others, made me realise there was more to intent than convenient and surface classifications. In addition, going through entries or articles on the Stanford Encyclopedia of Philosophy on intentions, intentionality, belief, desire, causal decision theory etc laid a better intellectual foundation for pondering a better way to view search intent than the conventional approach. This led to more presentations on BrightonSEO on topics such as intent graph, using causal inference to analyse search intent and belief representation as an important element in user search intent analysis.
After these presentations, I took a step backwards to try to figure out the practicality of my ideas on the importance of commonsense reasoning in better defining search intents for better search intelligence and mostly premised on ConceptNet relations. Reading articles from Lawrence Barsalou on “Concept and Meaning” and the “Situated nature of concepts” was reiterating the importance of context and outcome in understanding human intentions and concepts. I have also carried out more readings in the different perspectives on categorisation theory and its impact on category members and concept formation. I was still living in the theoretical world and wondering if there are any research or live projects that have brought some of these thought patterns to fruition.
I recently stumbled upon the amazing work by the Amazon Science team on creating COSMO, a large-scale commonsense generation for the e-commerce industry, and FolkScope focused on commonsense ecommerce discovery. In the research paper, they validated my argument in my BrightonSEO presentations that commonsense knowledge is useful in understanding user intentions. There is also the Intention knowledge graph project, powered by the Amazon m2 dataset. These utilised some relations from ConceptNet, mostly causal, temporal and conceptual. It is very refreshing to read that the collective intention knowledge graph of COSMO and FolkScope understands the importance of belief and desires in ascertaining and understanding intentions. As the consensus from these projects was that intentions were implicit and that events on the website like search, browse, purchase etc., were actions that are predicated on the communicated intent based on the belief the users have of the price, product attributes and the desired outcome.
It is rather refreshing to see that the combined intention knowledge graph has been an academic success, with a first-time acceptance of SIGMOD and has been implemented on the Amazon shopping app and is now live with the new Amazon shopping assistant, Rufus. This technology has been used in powering several solutions in Amazon, such as search relevance, session-based recommendation and search navigation. It also contributed to several hundred million dollars of revenue in 2023. Indeed, an academic delight and a commercial success.
This is indeed a validation of my earlier presentations and intellectual uneasiness of legacy systems and conventional assumptions in the search marketing field. COSMO and FolkScope have now laid the foundation and necessitated the need for a deep intention-first approach to search marketing challenges like user intent modelling, content gap analysis, search demand prediction, search journey mapping and a few more. Watch this space as SemanticGeek will answer the challenge with the right solution and platform in due course.