Interactive query expansion for professional search applications

I’m delighted to report that our paper ‘Interactive query expansion for professional search applications‘ has just been published in the journal Business Information Review. It’s been a long time in the making, primarily as the work was completed over an extended period of time involving many iterations, not all of which were successful (in the product development sense). But it does represent a concise summary of our work in investigating knowledge-based and distributional (word embedding) approaches to the generation of interactive query suggestions for professional search (which, I should point out, poses a qualitatively different and greater challenge than query suggestions for traditional keyword/web search). In fact, this paper represents an abridged version of the full results, since for reasons of space we were obliged to omit certain techniques that were less successful. For full details of those, please refer to our pre-print on arXiv.

Anyway, feedback so far has been very informative. Keep it coming in! New release coming soon, so happy to accommodate thoughts & suggestions. Abstract appended below.

Advertisement

Read Full Post »

Designing the Structured Search Experience: Rethinking the Query-Builder Paradigm

I’m pleased to report that our paper ‘Designing the Structured Search Experience: Rethinking the Query-Builder Paradigm‘ has finally come out in Weave: Journal of Library User Experience. It’s been a long time in the making, partly as Weave comes out only twice a year, but mostly as we wanted to include some key changes and updates from our latest work. What makes this paper different is that for the first time we’ve had the scope and mandate to explore the UX issues in some depth. Whereas in previous papers we’ve rather glossed over these things as ‘simply good design’ or ‘what UX practitioners are meant to just get on and do’, in this paper we took the opportunity to demonstrate how solving a challenge like this involves a level of nuance and complexity that may not be immediately apparent to the untrained eye, and moreover, that what you see in the deployed app is actually the culmination of significant design exploration and thinking – much of which ends up on the ‘cutting room floor’. Even just the algorithm for painting the iteratively nested containers (in alternating white/blue) was non-trivial, as you need to avoid edges where boxes meet at different level of nesting, i.e. resulting in a white on white or blue on blue boundary. Ironically tho, that bit of detail, interesting tho it is, didn’t make the final cut: you’ll have to wait for a subsequent blog post for that 🙂

Anyway, feedback so far has been very informative. Keep it coming in! New release coming soon, so happy to accommodate thoughts & suggestions. Abstract appended below.

Read Full Post »

NLP techniques for automated query suggestions

A key task in formulating effective search strategies is the identification of appropriate keywords and controlled vocabulary terms. Interactive features such as query expansion can play a key role in supporting these tasks. In this presentation we investigate a variety of methods for interactive query expansion based on manually curated resources (e.g. ontologies and terminologies) and on distributional methods (e.g. unsupervised machine learning). The results demonstrate the utility of distributional models and the value of using ngram order to optimise precision and recall. This work was due to be presented at the ISKO UK event Using knowledge organization to deliver content (which will hopefully be rescheduled later this year).

Read Full Post »

Advanced search needs advanced suggestions

Do you use search suggestions on Google? I’ll wager you do: they save time, help to minimise error-prone keystrokes, and even if not consciously processed, can give us the inspiration we need to form better queries than we’d originally contemplated.

But query suggestions, as we currently understand them, have their limitations. In particular, they are predicated on the assumption that most queries are composed of relatively short sequences of keywords, perhaps with some elementary linguistic structure. This is of course largely true for web search, but for many advanced or professional search applications, the assumption breaks down.

Read Full Post »