Lexxe, the natural language processing search engine also adds the power of clustering to provide results with subset cluster. Drill down into clusters lets the engine further select the links relevant with the context mentioned by the cluster.
The makers designed Lexxe with the aim to answer short queries by collecting content from the unstructured text floating on the Internet. The fetching part is done on the fly and leverages computational linguistics to exclude irrelevant content. Also, the clustering of results provides the user with options on various contexts of the query to further drill down.
Lexxe’s Phrase Recognition method allows the engine to understand if the key words are formed as one or more phrases and to linguistically extract the relevant factual information. Here are the details on the technology, which also include Part-of-speech Tagging, Parsing and Word Sense Disambiguation.
Question answering technology deployed in search engines are mostly database-driven. It means the answers are pre-prepared and/or manually checked. The answer is accurate, if matched with a question. But on the other hand, it is very restrictive to factual information, particular when questions have link verbs as main verbs, e.g. “be” (is, was, are, were, etc).
Furthermore, available question answering machines in search engines return long answers with the entire sentences, very often with useless, redundant and irrelevant information usually coming in in more than a dozen and resulting in more reading time. It indicates those question answering systems may know the answers are probably inside some sentences, but not sure which words actually make the exact answer.
The site works best for short queries, which seems an unreasonable condition to impose on users. If I were searching for some exact data, the idea employed is to always pack as much information in the query (and hope that the exact query occurs in some document). So, grabbing attention from mainstream engines may not immediately materialize for Lexxe. But again, Natural language processing is about giving the system time, just like nurturing human learning.
Also, the next generation engines seem to be making a shift to being experts on the query posed, rather than pointing in the direction of the possible results. It’s a question of whether this approach can pan out to encompass the whole swath of content on the web that’s ever growing.