About Lexxe
News and Press
Quintura Expert Tests Lexxe
Quintura search engine expert K. Zaytsev recently published an article about Lexxe Search Engine and some test results from his experiments in Quintura Tech Blog in Russian. The following is the English translation of his full article with comments from Lexxe.
Title: Natural Language Search Engine (Lexxe) in Alpha Version
Source: Quintura Tech Blog [in Russian]
Website: http://qtechblog.quintura.com/?p=5 [in Russian]
Author: K. Zaytsev [of quintura.com, a visual search engine]
Translator: Dr. Vadim Dedov
Date: June. 6th, 2006
[Translation]
N.B. The italic and bold parts are comments from Lexxe.
Search Engines are getting smarter every day, but some users expect too much from them. The concept of 'Search 2.0' aims at gradual introduction of intelligent systems where equal dialogue between the engine and human is possible (service 'question-answer'). The problem is that some users are getting disappointed in capability of search engines, if they fail to find what they after. In addition to basic cluster search, Lexxe (Australia) can answer short questions (up to 10 words) by way of using unstructured text and web resources. Its developers believe that Lexxe is 50% more accurate, relevant and efficient than other search engines. Introduced as search engine of the third generation, it uses technology of cybernetic linguistics and combines search by key word with capability to search by natural language. Its technology is based on language and depends on meaning. The process of automation does require on-line editors. In response to request, Lexxe determines its types/sub-types, where questions are determined by the question words in the beginning of a sentence (in the future they plan more flexible question types. Further, a question form is transformed into a statement form. The clusters (grouped by theme or by context) do not have hierarchic structure and their parallel use saves time for the user. On the basis of function for word determination, Lexxe determines addresses. Three types of results are shown on the screen: short answers, clusters and web search results, which are short summaries of the context. The aim of Lexxe is to extract one short, but exact, answer to the query from two respectable websites WordNet and Wiki without referring to location of possible answers. [Comment from Lexxe: that is not true. Lexxe tries to answer short questions by analyzing texts in webpages and retrieve them from sentences. Only when such effort could not return a confident answer, will Lexxe use WordNet and Wiki.]
On testing, the 'third generation search engine (Lexxe)' happened to be not always so smart. The answer to the question: When does Apocalypses occur? was 2005 by the 13th reference, which is in past tense. However, in half of the search query test cases, the Australian search engine was on top. The answer to the question: 'What is Quintura?' was 'visual search engine'. The answer to the question: 'Who will win the Soccer World Cup in Germany, was 'England'. Hopefully, the machine knows better. It is pity that there is no such service in Russian. [Comment from Lexxe: Lexxe's log showed that in late May 2006, there has been a series of tests from a single Russian site with 1124 test queries on Lexxe, among which 682 (60.67%) were questions and 442 were keywords (39.32%). Lexxe demonstrated its strength with half of the questions answered satifactorily. By studying the test, we have every reason to believe that this is a remarkable result of world class.]
|
