Enterprise search can trace its origins back to the early 1980s and for the last 40 years the research and development efforts have focused on improving relevancy, be it in the context of optimising precision or optimising recall. Much of the recent research using artificial intelligence and machine learning has the same objectives. Another recent development has been a significant increase in the number of connectors on offer to support federated search, with many vendors able to index 200 or more different applications. Most search vendor web sites are predominately about convincing customers that a solution to all indexing and query challenges is available off the shelf.

Enterprise search is different to all other enterprise database applications in that the user interacts with a computer to obtain a nominally best match to a query. This may take several sessions over an extended period of time as the user realises that the query needs to be recast, giving a new set of results to work through. The importance of the user interface in supporting this optimisation is critical to the success of a search session and yet very little attention has been made to the design and management of enterprise search user interfaces. Vendors are very keen to offer federated searching in multiple languages across both unstructured and structured content in almost any application or repository without explaining how the user is going to cope with a very complex user interface.

It does not help when the accepted wisdom (without any evidence) is that enterprise search should be intuitive and require no training.

We have reached the stage where all enterprise search software is based around a well-developed set of algorithms (for example BM25) and yet users are still dissatisfied with their organisation’s search applications. In my experience the issue is not about what might be described as the technical performance of the search application but of the difficulty of using an interface which is almost always generic and not optimised for specific tasks. Research into professional search has shown that the ways in which lawyers, clinicians, patent agents and recruitment managers use the features of a search application are different. For example, clinicians make extensive use of synonyms (heart cf. cardiac) and acronyms, neither of which are of value to lawyers.

It does not matter how sophisticated the enterprise search technology is in terms of the features and functionality it offers. What matters is if the user can make an informed judgement about which piece of content presented in the list of results best serves their information requirement, reinforces their trust in the application and maintains the highest possible level of overall search satisfaction.

How the results are presented to the user is therefore critical in enabling the process of relevance judgement.

Result scanning

We tend to talk very glibly about scanning a list of results from a search without for one moment considering what this action involves. The speed at which the results can be scanned and appreciated in terms of their potential relevance varies from searcher to searcher. The concept of perceptual speed is usually totally ignored. This is a cognitive ability that determines the speed in comparing figures or symbols, scanning to find figures or symbols, or carrying out other tasks involving visual perception.

It is important not to confuse perceptual speed with readability. Perceptual speed relates to the ability of the searcher to make out words and other information elements. Readability is about the comprehension of those elements in the process of extracting information and knowledge.

Perceptual speed is not easy to measure but the impact on the search user can be quite dramatic. Users who are dyslexic face particular problems in scanning at speed. An outcome might be that the process of scanning is slow enough for a user to give up on the process after a few pages of results, and so not find all the relevant items.

Result reviewing

The next step is to look at each result and decide whether it is relevant enough for us to take the time to click on it and open up the content item. Simple! Or is it? In common with many aspects of enterprise search there seems to be no research on how snippet length and design enable an informed decision on relevance to be made.

There is some (but arguably not enough) research on snippets for web search queries but in general these snippets are linking to a web page which can be scanned and assessed reasonably quickly. In enterprise search the content item could be several hundred pages long and it may be far from obvious where the relevant information (according to the ranking algorithm) is to be found.

There are three fundamental ways of generating a snippet:

  • present the query term in a text sequence that should provide enough context for the relevance to be assessed
  • create a computer-generated summary of the content item
  • reproduce the first few lines of an abstract (see Google Scholar for examples).

Some search application vendors provide a thumbnail of a page that contains the query term, but the accessibility problems arising from having to view a small image displayed as the result of very precise mouse control are ignored.

The duality of search use

Enterprise search is almost certainly used by the majority of employees in an organisation. Most of the queries will be sub-critical, in that search is a convenient way of tracking down specific items of information but not the sole way of doing so. However there will be many employees who will be using the application to enable business-critical decisions to be made where a failure to locate the information needed could put the organisation at risk. It is also likely that in these situations the employee cannot be certain of which applications (and that is a deliberate plural) they need to search through in order first to find a range of relevant information and then have to integrate the results and synthesise an outcome to make as informed a decision as possible.

Over the last two years there has been a significant increase in the amount of academic research being undertaken into the search process. David Maxwell’s thesis on information foraging and stopping distances makes a significant contribution to understanding the cognitive processes behind reviewing search results.

The screenshot below comes from research carried out by Hugo Huurdeman and his colleagues at the University of Amsterdam and the University of Nottingham. SearchAssist integrates both a results display and a range of search support features to support a multi-stage query and result process.

Screenshot SearchAssist. Left column (1, 2, 3): control features. Middle (4): input and informational features. Right column (5, 6): personalizable features. (7): task bar.
(From H.C. Huurdeman Supporting the complex dynamics of the information seeking process PhD thesis University of Amsterdam 2018 ISBN9789082169508 https://hdl.handle.net/11245.1/1e3bf31a-0833-4ead-a00c-4cb1399d0216.)

In addition there will be employees who will be using the enterprise search application on a regular basis, perhaps several times a day, and will become conversant with even a very complex user interface. It is quite probable that they will wish to optimise the layout for different but related tasks. Research by Tony Russell-Rose, Jon Chamberlain and Leif Azzopardi into the search interface requirements of professional searchers indicates some important differences between the way in which patent agents, recruitment agents, lawyers and healthcare professionals use elements of the user interface.

Customising the user interface

The concept of customisable user interfaces has been under consideration for a number of years. It is well worth reading the outcomes of the Khresmoi project conducted in 2013/2014 with the objective of developing a multilingual multimodal search and access system for biomedical information and documents. This was achieved by:

  • effective automated information extraction from biomedical documents, including improvements using manual annotation and active learning, and automated estimation of the level of trust and target user expertise
  • automated analysis and indexing for medical images in 2D (X-Rays) and 3D (MRI, CT)
  • linking information extracted from unstructured or semi-structured biomedical texts and images to structured information in knowledge bases
  • support of cross-language search, including multilingual queries, and returning machine-translated pertinent excerpts
  • adaptive user interfaces to assist in formulating queries and interacting with search results.

This research project led to the development of ContextFlow as a search application specifically designed for radiologists.

Sinequa, a French enterprise search software company, offers a range of customisable user interfaces, predominantly for the pharmaceutical and manufacturing sectors. The image below is a user interface for research scientists searching both internal and external sources of information. This interface can be customised by the search users.

© Sinequa. Used with permission.

A tipping point for enterprise search?

The technology behind enterprise search dates back to the early 1980s, with a major leap in functionality with the adoption of the BM25 ranking model in the mid-1990s. The BM25 model has gone through a number of variants and is now complemented with knowledge graphs and AI/ML routines. However the perceptual impact of these developments on users of enterprise search applications is arguably increasingly limited because of the inherent issues of the variability of content in enterprise collections and the range of intents of users. It is becoming very difficult for enterprise search vendors to differentiate their product offerings!

There now seems to be a substantial opportunity to offer search user interfaces that are optimised for specific tasks and/or capable of customisation by users. This approach is already being used by Sinequa in a number of areas, notably searching through clinical trials data and in product development applications. The impact of these developments on the search user is of course very visible, and the benefits in terms of productivity, innovation and speed of response to customers can be much more easily quantified than in text search. As a result it is easier for vendors to make a business case to prospective customers and to differentiate their offerings from competitors.

The result is likely to be that vendors and integrators will quickly appreciate the benefits of providing a much higher level of enterprise search experience than they have over the last 40 years.

This chapter was originally published in Search Insights 2021. Download the report for free here:

Search Insights 2021 by The Search Network
Search Insights 2021 – ENGLISH
Search Insights 2021 – FRENCH TRANSLATION

Leave a Reply