At Qualcomm, they have created Enterprise Centers of Excellence (CoE), including Search, Content Management, Collaboration and Desktop, combining leaders from IT, Engineering, Program Management, Finance, and more. The Search COE has succeeded in providing a central organizing point for search, muting search technology affinity wars, and increasing awareness of search capabilites leading to increased demand for specialized tools.
Mark Livingstone of Qualcomm and Miles Kehoe of New Idea Engineering will be presenting about this CoE experiences at the Enterprise Search Summit, and we'll have an open discussion at ESSF, moderated by Lynda Moulton, where experts and conference participants can share experiences, good and bad.
The Enterprise Search Summit Fall will be held November 1 - 3, 2011, in Washington DC -- I hope to learn from you there. For a $200 discount, use the code at the online registration page.
New to our lineup is Greg Nudelman, author of this year's best book, Designing Search: UX Strategies for eCommerce Success. He's been involved in several mobile search interfaces, and will present Ubiquitous Enterprise Search: New Design Approaches for Mobile and Tablet -- this is going to be good! Register for the conference before October 7 to get the early-bird discount.
BTW, I've been so busy with this conference and a large contract with a giant healthcare system's intranet search that I haven't been very responsive, and I apologize. If you need something from me, please remind me by commenting here or sending email, don't be shy!
Good intro to ecommerce search based on significant online retail experience.
Good overview of enterprise search issues, contrasting unstructured search with RDBMS processes and discussing issues of granularity and simplicity.
Traces the algorithmic relevance problems in search on a textual database of authors and books. This chapter shows how to use the Solr DisMax functions to add extra relevance weight for matches on canonical names and alludes to adding weight for titles matches -- well-known heuristics for getting more useful search results.
To extract assertions for question-answering systems, the ReVerb package implements the Open Information Extraction (Open IE) theory. Without needing complex training examples and domain-specific words, this simply extracts relationships, such as "apple has phone". It's open-source, along with the extracted sets from wikipedia and TReC's ClueWeb09 corpus.
Competing with sharepoint search on near-real-time index updates, faceted refinements, faster crawling, default configuration is easier to maintain, more extensible in pipeline, etc. Price is about $14,000 per server, no need to buy additional SharePoint licenses.
Talks about a 2011 analyst report on Enterprise Search engines and trends like pervasive search, search within business processes, pro-active search managers, using search to find workplace patterns, demand for transparency and user control,
Informal conversation starts with AT&T's CMS and goes on to talk about using Solr for catalog search on the site. It started as a bit of a skunkworks, and takes very little maintenance and few resources.
Excellent introduction to the default Solr spelling checker processes. The test information is particularly useful, including data sets with natural human-made spelling errors. There's a follow-up that's also very useful.
Lucid Imagination is using the open source enterprise connector famework with their LucidWorks Enterprise distribution of Solr/Lucene.
Thinking about relationships with search engine vendors, market niches, diverse information requirements, and many more issues in managing search.
Starts with Solr/Lucene and adds autosuggest for faceted metadata, geospatial filtersing, spellchecking, search sessions, integration with external resources and databases, related documents, simple bar, pie and column charts, integration with Quantum4D for visualization.
Danny Sullivan's detailed look at how Google's auto-complete and instant search, which seems to be populated by a combination of many recent searches and long-term frequent searches. They have an inconsistent policy on removing negative terms from the suggested list. He thinks they should just remove them without fuss.
These examples emphasize how enterprises and web sites should keep very strict control over their autocomplete (and spelling) suggestions, rather than rely on technology.
In e-commerce, it's much better to find something than nothing. Spellchecking and other tools for inexact matching can improve results for the long tail of unique and near-unique queries. Exorbyte reports increases of conversion from 5% to 20% - at least somewhat related to their error tolerance.
We particularly need more talks on open-source search, scaling way up, usability testing, and multimedia searching (images, audio, video).
Enterprise Search Summit, Fall 2010: Call for Presentations
( Collapse )
Enterprise Search Summit 2011, Program chair Avi Rappoport
Describes Attivio's combination of unstructured and structured data for a new kind of BI. ... entity data and the sentiment analysis and other analysis techniques become part of the meta data for the unstructured data. This allows the unstructured data to be presented as structured data. You can do relational joins between the structured and unstructured data to answer questions like: “Find all negative product reviews that mention both the iPad and our top-20 best sellers.”
Consumers using site search browse about 11 more pages per visit than shoppers who don’t, and remain on the e-commerce site nearly 13 minutes longer.
"IDOL Universal Search provides users with a simple, personalized search experience tailored specifically to their unique requirements. In many typical environments, content may be spread across disparate internal data repositories and systems as well as external content sources and engines. Users are forced to conduct multiple, repetitive searches or risk missing crucial information. With Universal Search, a single query quickly federates unified results across systems in an intuitive, easy-to-navigate Web interface. Users get the big picture quickly through advanced visualization and dynamic categorization for a faster, more thorough exploration of results."
Results from a usability test - only 1% of the time did users change search strategy. So search engines need to concentrate on improving reliability of simple search results.
Tony Russell-Rose does a great job identifying various models of faceted search interactions, including the relations between the facets, Smart Dead Ends, the relative values of "instant update" and "two-stage" matching and user choice, multi-select AND vs. OR, and interstitial pages. Very valuable analysis and naming.
Concept searching uses lemmatized views of words to offer valuable autocomplete suggestions.