Telcordia Latent Semantic Indexing (LSI)
Demos
[LSI HOME |
EXEC SUMMARY |
DEMOS |
PAPERS |
CONTACT US]
This page describes two opportunities to see LSI in action. The first
allows you to experiment with LSI by entering queries against some document
sets on this site. The second allows you to see LSI produced recommendations
at a Telcordia e-commerce web site.
LSI Queries
This site contains LSI indexes from 3 document sets: A Canadian Parliament
cross-language set, the Bible, and a collection of sailing related
web pages. Clicking on this LSI
Query link will take you to the query page. You may like to try out
the following two examples to see a sample of the power of LSI.
- Cross-language Example
- Next to the "best matching" button select "Terms & Documents"
- Make sure that "Searching in" is set to the
"Canadian_Parliament" document set
- Enter the term "conservatives" (without quotes) as the Keyword Query,
then click on "Submit Query" button.
- You will get back a list of all documents and terms that are
nearest neighbors to the vector for "conservatives" in the LSI vector space.
Notice that the documents in French were retrieved, even though the
word "conservatives" does not appear in the French documents!
- Higher-Dimensional IR Example
- Make sure that "Searching in" is set to the "Bible_By_Verse"
document set
- Input as "Keyword Query" the term "manna" (without quotes),
then click on
the "Submit Query" button.
- You will get back a list of Bible verses with the term
"manna" as well as others with terms related to "food",
"eating", "bread", etc.
- Note that this search was
performed in an LSI vector space of 312 dimensions.
- You can also experiment with the "relevance feedback"
feature of LSI by clicking-on the check-boxes in front of documents you found
relevant in the Matching Documents return list. After clicking on
the related documents check-boxes, again click on the "Submit Query" button.
This sends these documents as a query to the LSI search engine, i.e., find more
documents like these.
LSI Recommendations
The Telcordia Information SuperStore is a site where customers can
purchase Telcordia products, primarily documents, as well as training and
services. An LSI scaling was run over the abstracts associated with most of
the products in the Store. When a customer
adds an item to a shopping basket, the 4 nearest neighbors in the
LSI space to the abstract of the new basket item
are returned as recommendations of other items the customer may
find useful. An empirical study of the effectiveness of LSI recommendations
was reported at CHI 2000. One result of
that study was that, "8% of the items added to baskets in orders were
recommended items and this resulted in a 7% increase in revenue."
To see this functionality, click on the
Telcordia
Information SuperStore link and try the following example.
- In the Navigation Bar at the top of the page, click on Search
- In the Keywords form item, type in the term: GR-1063,
which will allow you to find that document, and click on Submit Search
- On the Search Results page, click on the link to GR-1063
- On the Full Description page, click on Add Item to Shopping
Basket at the bottom of the page
- On the Shopping Basket page, notice the section at the bottom,
entitled Related Items You May Wish To Consider. These are the
result of an LSI search to retrieve the 4 nearest documents to the one
in the basket