The 'word of mouth' knowledge that swirls around organisations as people swill their coffee seems to have become a prime target for European software research and development outfits. The development and use of intelligent agents to build up knowledge communities at the BT Research Labs is one example of this trend. Another is the software developed by Orbital Software that matches knowledge seekers to knowledge holders (see issues 143 and 144 ofIWR, respectively).
On the continent similar things are happening at the Xerox Research Centre Europe (XRCE) in Grenoble. Knowledge Pump is the Xerox tool closest in conception to the BT and Orbital technologies. It is one of a suite of products that includes Knowledge Broker, the meta search engine currently being brought to the library market.
Christer Fernstr”m is Area Manager of Co-ordination Technologies at XRCE in Grenoble. He says that these tools are just parts of a set of components "that will help user organisations build knowledge management solutions".
Last year Xerox spent $1.1 billion ? about 6 per cent of its revenue ? on R&D. Its European presence features two research laboratories ? one in Cambridge and one in Grenoble ? which complement the work of the Palo Alto Research Center (PARC) in California. The two European sites are supported by a European development group that facilitates technology transfer from research into products. A notable feature of Xerox's R&D is its use of anthropologists to study human-technology interaction; thereby attempting to ensure that the technologies emerging actually address cultural issues.
Dr Fernstr”m comments: "The philosophy of Knowledge Pump is to automate 'word of mouth' knowledge transfer." While the tool is still in development, there has been a prototype in use at Grenoble for the last half year, and this has recently been transferred into the development group.
The core technology behind Knowledge Pump is called 'community-centred collaborative filtering'. Its action is relatively straightforward. 'Recommenders' classify items using an agreed scheme of communities of interest. Knowledge Pump then uses a proprietory algorithm to perform automated collaborative filtering, and this statistical algorithm makes recommendations based on correlations between personal profiles.
So, how do you use Knowledge Pump? If you come across an interesting piece of information you tell the Pump about it and you channel it towards the Knowledge Pump-defined community within your organisation that you think will find it of most interest. Within Xerox, Dr Fernstr”m gives examples of such communities being 'those interested in specific technologies (distributed systems, Java, e-commerce)' or 'those interested in marketing issues'. The locator of a piece of information gives the resource a 'relevance' rating and can also add annotations to say why it is of interest. As a user of Knowledge Pump you get "a hit list of potentially interesting things to look at", according to Dr Fernstr”m, who added, "I subscribe to four or five communities and get three or four recommendations from each community every day."
He maintains that the system is highly tuned to the knowledge needs of each particular individual. "Knowledge Pump uses users' areas of interest and feedback in the form of qualitative judgements. For example, Tim and Tom may share a common interest in the Java language, and their profiles built up from keywords of reading material may have a very good match. However, one of them is a development manager and is interested in compatibility issues whereas the other is a developer, interested in new event models. Although they share an interest and have matching keyword profiles, they would probably not value articles in the same way. By giving their own feedback to Knowledge Pump, their profiles will diverge ... and Knowledge Pump will stop cross-forwarding articles between them on the subject of Java."
This all sounds nice in theory, but what about the cultural issues that arise from deploying this technology? One set of concerns arises when the question of privacy is considered. Dr Fernstr”m admits that this is "something we need to look into more seriously". For example, what if a knowledge item is recommended by your superior but you think it is next to worthless? If you refrain from telling the Pump this, your profile begins to wither ? like the picture of Dorian Gray. Of course, you can tell the Pump things confidentially; but then you are hardly participating fully in a community of interest.
Another set of issues has to do with incentives. Why should you contribute information to the Pump and why should you add judgment and commentary to that information? Dr Fernstr”m concedes: "It does add yet another burden ... many people [at Xerox] were keen to get the recommendations, fewer people were keen to provide new items and even fewer wanted to provide elaborate comments." And so they have inaugurated an economy of 'chits'. When you retrieve something from the Pump you pay a chit, when you provide something you get a quarter chit and when other people use what you have recommended you get stronger reimbursement ? royalties, in effect that "show people what they do is useful".
An interesting aspect of the use of this technology lies in its constitution of communities of practice on the basis of users' profiles. In other words, the technology creates the communities. Without the technology you would have no chance of belonging to a virtual community within a global organisation that was involved in a topic that you had no idea either existed or would be interesting. This area is a 'project within the project', and is being explored by anthropologists at Xerox PARC in California. Dr Fernstr”m says that the project's researchers "want to understand patterns in organisations that go across functional areas, and to understand what these virtual communities might be".
Arguably, it is this international perspective that gives Xerox's R&D something of an edge over similar projects. The organisation itself can command resources in the States and in Europe and field trial its tools with different language communities.
Knowledge Broker, for example, has been tried out at the French Telecom Centre and at the Bayerische Staatsbibliothek in Munich. Due for commercial release about now, it is a meta search engine that sits on top of other search engines and searches heterogenous sources. According to Xerox, in comparison with similar information retrieval technologies, it offers a range of distinctive features that allow complex queries, dynamic refinement and multiple queries. The system uses Brokers and 'wrappers'. A Broker manages one request from one user (many users can submit many requests at the same time). The wrappers recover structure from raw information, such as Web pages, and enable the user, who will invariably be an information professional, to constrain searches to specific databases. The system boasts portability: the client and servers are implemented in Java, and create no data replication or unwieldy local indexes. Moreover, since it integrates sources and handles access rights, Knowledge Broker seems promising in an information environment where information from outside and inside sources needs to be blended, and legacy databases need to be connected with an up-to-date user interface.
The drive to create Knowledge Broker at Xerox came from the need of its R&D personnel to access multiple information sources in one go. The initial work at the labs used constraint-based programming technologies, and, about two and a half years ago, resulted in a prototype that could be trialled with some early adopters. The query language of Knowledge Broker follows a syntax based on Boolean connectors.
One significant early adopter is the Bayerische Staatsbibliothek. One of its librarians, Dr Astrid Schoger, tells us that Knowledge Broker is trained on six information sources: the Research Libraries Union Catalogue (RLIN), the Bavarian Libraries Union Catalogue,MEDLINE, VD17 (a catalogue of seventeenth-century German language texts), and the AltaVista and Excite Web indexes. In Dr Schoger's view, tools like Knowledge Broker "offer a homogeneous view of heterogeneous data repositories". She goes on to make the obvious, but key, point that a lot of time can be saved by being able to "search many data repositories at once". She also points out, however, that "end users have to be aware that these benefits go hand-in-hand with a loss of information in comparison with the results which direct searches in all repositories would produce".
Asked about the difference the increasing use of such tools will make to the job of the information professional, Astrid Schoger commented: "libraries have to think more globally". She added that librarians still have an actively integrative role in that context because "intelligent software alone will not overcome the heterogeneity of information" that exists internationally.
Christer Fernstr”m makes similarly internationalist and integrative points in highlighting the fact that Knowledge Pump and Knowledge Broker belong to a suite of tools including linguistic processing systems. "Using linguistic components", he says, "will support users in a multi-lingual environment such as Europe and provide semantic analysis and processes such as summarisation, synonym matching and so on."
The most interesting aspect of the deployment of these tools may well prove to be their sub-serving and creation of virtual communities ? or indeed precisely their failure to do just that.
reader comments