In December I spoke on a panel at SES Chicago about how Semantic Search will change our lives. I wanted to do a blog post about what some of the speakers said.
Semantic Search means various things. To several of the speakers on this panel it means search based on recovery of the detailed linguistic meaning of the query and target document base. To others it means recovery of “sentiment” or evaluative language in customer reviews of vendors such as restaurants. To others it means search in semantically tagged fields or structured data in a relational database.
Several of the speakers in the panel agreed that a radical shift in search in which meaning-based indices are searched, rather than pattern-based indices, is going to make the next big improvement in search. Similarly, ads will be placed based on meaning. As Tim Musgrove of TextDigger said, it’s silly for various vendors with different types of businesses all to fight over the keyword “palm”. The word (or pattern) is relevant to palm trees, palm pilots, or palms of the hand (say for a sporting-goods vendor). If linguistic semantic search replaced string search, there would be three different keywords (for different meanings of “palm”).
The Powerset speaker (Scott Prevost) agreed. He explained that Powerset linguistic processing enables their search engine to distinguish “who did what to whom”. This avoids false hits due to argument structure, so a query about “Who did Merrill-Lynch acquire” does not retrieve a document about Bank of American acquiring Merrill-Lynch. Also, Powerset’s linguistic semantics enables their search engine to find facts about query entities in free text, and use those facts to enhance search result displays.
Larry Cornett of Yahoo! described the use of structured data along with free text to improve the effectiveness of retrieval results. Developers are allowed to add structured data and create applications that modify and augment the display on the Yahoo! results page. For example, a restaurant review developer could show ratings, reviews, address, and other information most users want about a restaurant, directly on the results page. This type of semantics dovetails with Web 3.0 tagging in that structured semantic information is introduced by hand through knowledge engineering.
Nadaraju Bandaru of Boorah! explained “sentiment extraction”, where reviews of offerings such as restaurants are rated on sentiment scales to decide whether they are “boo” or “rah” ratings. These results are displayed to help users quickly decide upon their choice. Sentiment scales are determined by the semantics of words in the reviews such as “yummy”, “delicious”, “nice”, etc. Boorah! also illustrates the advantage of focusing on a vertical to narrow the range of meanings, improving precision.
This panel suggests that there is significant interest in improving the search experience with semantics, especially deep linguistic semantics a la Cognition and Powerset.