Innovation, compliance, learning – important roles for exploratory search

by | Feb 21, 2018 | Search

There is a substantial amount of promotion of the benefits of precise personalized search at present, and I have commented on some of opportunities and issues in an earlier post. Very little is being said about exploratory search. Let me start by quoting from a very recent paper from Di Sciasco, Sabol and Veas in which they present a novel user interface specifically to support exploratory search.

“It is clear that for an exploratory engine it is not enough to know just what we previously searched for; it also needs to know what we did with what we found. Was it categorized, was it good, and did it lead to another topic and another search? Indeed, collecting information about a new topic is rarely solved with a single query. It is instead a discovery process involving several queries intermingled with extensive reading of retrieved resources. Query terms are refined and reformulated as results are explored. Interesting results are collected along each step, but the connections between them exist only at a cognitive level in the user. The effort and time are mostly spent in careful reading and acquiring information from search results (often reading titles, abstracts, and text summaries) and in formulating new concepts to continue exploring.”

There is a frequent requirement to explore the information resources of the organisation to learn about a topic, to look for innovative approaches to solving a problem or to respond to a compliance requirement. The focus shifts from precision to recall. In an exploratory search the user has no preconceived view of what they will find. If they did it would be a precision search. Search needs to support a dialogue between the user and the machine, checking ‘Did you mean?’ and offering suggestions for query terms. It will also present a set of filters and facets that enable the user to go off in different directions and still be able to return to the starting point.

Closely related to exploration is that of serendipitous discovery. Interestingly (at least to me) is that the word was first used as long ago as 1754. Replicating the neural connections needed to make serendipitous discovery are (but I will probably be wrong!) some way in the future. In the mean time we can support serendipity by custom-designed search interfaces that encourage us to head off down a path in which we make individual decisions as we see best. There are strong indications that exploratory search can enhance serendipity and innovation. If you want to read a good example of search innovation go back to the research paper from which the opening quote comes from. The authors have developed a very interesting user interface to enhance the process of exploratory search. That raises the issue that most search user interfaces belong to the ‘one size fits all’ category – there is huge scope for improvement here.

One of the challenges of ‘recall’ is that you have no way of knowing if indeed all relevant information has been presented. Again you have to trust the technology. I note with interest that Attivio is committing “to deliver ALL (Attivio caps) the relevant information”. 100% recall! The relationship between precision and recall performance is a complex one and way beyond this post to consider in detail. Precision-recall diagrams are very helpful in optimizing search performance and it is important to consider them together. For now I would like to pose a question. How confident would you be that the algorithms being used for precise personalized searchwill not degrade recall performance? A user may post a query for which they are expecting a precise answer. If the answer they get diverges significantly from their expectation do they repeat the same query and hope that the application now works in exploratory mode, or do they have to change the query? Or do something else?

Without doubt enterprise search technology is going through a period of rapid development in terms of functionality. AI, machine learning, natural language processing (NLP), content analytics, topic extraction and visualization are all going to expand the capabilities of enterprise search. That does not mean to say that enterprise search is dead. The fundamental elements of the technology create a platform for this development. The problem that organisations are going to face in evaluating the impact of these technologies is that the process from the initial definition of user requirements through to the global launch of the application could take at least 12 months and more probably 18 months. Testing out novel technologies at a ‘proof of concept’ stage is not easy because the last thing you want to do for a PoC is carry out a full index. In addition, testing out these technological innovations requires users to undertake a reasonably substantial amount of searching to build up the weak signal collection. Try asking a vendor about what they would regard as a number for ‘reasonably substantial’ to validate the technology at a PoC stage.

The use of these novel technologies could be of substantial value to an organisation and I hope that many will take the required leap of faith and adopt them. However first you will need to come to grips with information quality and you will need to have a search team with the skills and resources to support the definition, selection, testing and implementation of these next-generation applications. Add together information that has consistent quality, a good search team and good technology and you will transform the performance of your organisation.

Martin White