BOOK AND SOFTWARE REVIEWS


Susan E. Feldman The answer machine. Morgan & Claypool Publishers, 2012. 137 pp. [PDF file] ISBN 9781608459353 Single copy of file $20.00, paper copy $35.00 (Synthesis Lectures On Information Concepts, Retrieval, and Services, Vol 4 No. 3).


I need to start this review out with a declaration of a bit of a conflict of interest. I recently became the co-editor of what some people might regard as a rival journal to Gary Marchionini’s synthesis lecture series; namely Foundations and Trends® inInformation Retrieval (FnTIR). Both of these journals have been producing monographs addressing topics in the field of IR for the past few years. Gary is doing a great job getting prominent people to write for him. Reviewing Susan Feldman’s piece was my first opportunity to read one of the Synthesis Lectures in detail. After reading this, however, I realise that the conflict between our two publications isn’t as strong as I thought, as, if Feldman’s piece is typical, then the lectures are more detailed opinion pieces, which contrasts with the literature review articles produced by FnTIR.

Feldman seems to have been inspired to write this piece by the success of IBM’s Watson system that beat two human champions from the long running US quiz show, Jeopardy! She sees Watson as an indicator of what is to come: no longer will we use search engines to retrieve documents, we will have computer systems that give us answers to our questions.Feldman sets out her case by first making clear who she sees as her audience (non-techy people) and then proceeds to explain how queries to search engines are different from requests for information elicited from people (Ch. 2). She follows this with a description of legacy search engines (Ch. 3) and more detail on more recent developments (Ch. 4). I am a hard core techy, so there wasn’t really anything in the explanation that was new to me, but I found the summary to be clear.

What riled me in these sections was the odd excessive generalisation, occasional error and a lack of references. Reading Chapter 3 made me wonder how fully up to date Feldman was with the underlying technologies of search; however, the following chapter made me more confident in her mastery of the field. A general lack of references,was annoying, however. For example Feldman states that in the 1970s, Spärck Jones created a document similarity scheme based on TF/IDF(term frequency/inverse document frequency). That was news to me, Spärck Jones invented IDF weights in the 1970s and also worked with Steve Robertson on very popular ranking scheme in that same decade, but it didn’t incorporate TF weights, so which paper is being referred to here? Probably more annoying were examples such as this; Feldman at one point wrote "Studies show that online buyers will not buy as often from Web sites that are not in their native language". I would really like to see those references and read up on them some more. But, as so often happens in this document, I’m just going to have to find them myself.

The next chapter (Ch. 5), Feldman puts a lot of time into describing the needs of users across many different applications. Feldman does a good job of detailing a wide range of search request types, outlining the different nuances of need. Showing well how search is far from being a solved problem, Feldman shoots through the current developments in search by describing different means of learning user intent in a query based on analytics as well as methods for determining context and displaying results in novel forms (Ch. 6). After that Feldman’s vision of answer machines is laid out (Ch. 7) explaining how she sees the creation of systems like IBM’s Watson will come about.

I liked the way that Feldman sets out her case; I’m pretty sure that she is right about how search is going to evolve towards finding answers. Feldman is not the only person to describe these ideas in recent years, Croft for example wrote on the importance of providing "the most appropriate answer for a user's information need". The book is clearly for non-techies; those who keep up to date with research developments in information retrieval as well as question answering systems will find little new, although Chapter 5’s description of different types of need is a good read. If you are a non-techy who wants to top up on how search engines work then this document will help you out. If you need to be persuaded of the case for answer machines this is for you too.

I just wish there had been a few more references.

Professor Mark Sanderson
RMIT University
Melbourne
Australia