AskWiki

from Wikipedia, the free encyclopedia
AskWiki

Logo from AskWiki
Basic data

developer Telekom Innovation Laboratories
Current  version 2.2
operating system Android 2.1 or higher
category Books & Reference Works
askwiki.de (no longer available)

AskWiki was as Android - App available semantic search engine .

The app tried to meaningfully interpret a linguistic request and find a suitable answer. The application was designed for several languages ​​and used the respective Wikipedia (in German , English , Czech ) as a knowledge base .

functionality

Unlike an ordinary search engine , the main goal is not to find facts available on the Internet or to search for a suitable Wikipedia article using search strategies, but rather to process facts using a specific algorithm . AskWiki is therefore a semantic search engine . Possible questions could be:

The results are responses generated by the app that are derived from information contained in a Wikipedia article. The answers are either displayed in text form or reproduced through synthetic speech .

The processing process is divided into the following steps:

  1. The user request is recorded. This is done either by the speech recognition of Google in the Android system or by a direct input in a text box, wherein the language used by the general language setting of the Android device is aimed.
  2. Unnecessary words are removed from the input (the, the, that, who, where, when, etc. or their counterparts in the other languages) and word sequences are then systematically extracted and transmitted to the Wikipedia server. This happens until either the server delivers an article or all phrases have been tried.
  3. The item found is then analyzed in more detail. If there is an adjective when entering it, such as "old", an equivalent is searched for either in a table or in a chapter heading. If a match is found, the corresponding value field in the table or the first sentence of the chapter is used as the answer, otherwise the first sentence of the entire article.
  4. The answer is output in text form or by speech synthesis . In addition, images from the article or a link to the entire article are available.

The app runs exclusively on the smartphone, which means it does not need a connection to a server, apart from the Wikipedia server and possibly the server for Google speech recognition. No dictionaries are used to identify Wikipedia articles and property names. An ontology is also not used. A synonym vocabulary, which the user can expand himself, is used to search for correspondences for adjectives.

Problems

When searching for the title of the appropriate Wikipedia article, various spellings are tried out and an attempt is made to automatically suppress words that are part of the natural question but are not needed to answer it. For example, the question “Who are Einstein's ancestors?” Is changed to “Einstein's ancestors”. But of course it is not always possible to find the right article; The user may have to specify the term more precisely, especially in the case of ambiguous terms. For the question “Who is in Nashville?” The city “Nashville” would be matched and there is no meaningful answer to the question. With the request “Nashville Film”, this request is automatically converted to “Nashville (Film)” and the relevant Wikipedia article with the actors is displayed.

The user's search query should follow this scheme as far as possible so that any errors in speech recognition do not have a negative effect on the further processing steps. If, for example, in the above example, “Who are the ancestors of Einstein?”, The speech recognition returns “Who are the Sinti ancestors of Einstein?”, Then the intended result would not be delivered. The direct voice query “Einstein's ancestors” is more likely to be correctly recognized and processed with voice recognition.

So far, semantic relationships have only partially been explicitly stored in Wikipedia, for example in the information boxes of an article, so that the information in the app must largely be made accessible through text analysis. The search for answers would be easier if Wikipedia contained such relationships directly (see Semantic MediaWiki ), i.e. instead of the text "Berlin has 3,393,933 inhabitants", better "Berlin has [[has population :: 3393933]] inhabitants". Despite its ability to interpret natural language, AskWiki does not provide meaningful answers to many search queries. At least the appropriate Wikipedia article may be found and displayed.

Comparison with search engines

Compared to search engines, AskWiki gives you direct answers. For example, to the question “How old is René Obermann?” The answer is given directly in years; If you click on the hyperlink , you will find the date of birth at the corresponding point in the article on René Obermann.

Due to the complexity of the calculations, natural language queries are usually slower than with a search engine (you can virtually follow the various processes live in the app), but the question is then answered directly and the answer is not found behind you, as with a conventional search engine Hyperlink . In addition, the database of Wikipedia is not as large as that of an Internet search engine.

development

AskWiki was developed as a technology demonstrator by Telekom Innovation Laboratories and officially launched on January 27, 2012. The aim is to evaluate the experience and to transfer the methodology used to other fields of application. A year later, Ford built AskWiki into its AppLink or its SYNC AppLink mobile application catalog. The last known version was version 2.2 in May 2013.

At the end of 2013, the development was stopped and the app was taken off the market.

literature

  • Felix Burkhardt and Jianshen Zhou: "AskWiki": Shallow Semantic Processing to Query Wikipedia (PDF; 572 kB), in Proceedings of the 20th European Signal Processing Conference (EUSIPCO), August 27-31, 2012.
  • Max Völkel, Markus Krötzsch, Denny Vrandečić, Heiko Haller, Rudi Studer: Semantic Wikipedia (PDF; 568 kB), in Proceedings of the 15th international conference on World Wide Web, WWW 2006, Edinburgh, Scotland, May 23-26, 2006.
  • Research and Design Innovations for Mobile User Experience , Example 3: Field-Testing an App Before it is Released: AskWiki, p. 86 [2]

Individual evidence

  1. ↑ [1]
  2. ↑ W&V: Telekoms Speech Recognition App: Does AskWiki know Steve Jobs better than Siri? In: wuv.de. Retrieved October 20, 2013 .
  3. ↑ AskWiki - voice search for Wikipedia - GIGA. In: giga.de. Retrieved October 20, 2013 .
  4. ↑ Search engines are finally learning to think - news.ORF.at. In: orf.at. Retrieved October 20, 2013 .
  5. ↑ Deutsche Telekom: Forgotten about an answer? Telekom app AskWiki leaves no question unanswered. In: telekom.com. Retrieved October 20, 2013 .
  6. ↑ Ford brings AppLink Generation 2.0 and a SYNC AppLink Mobile Application Catalog to Europe
  7. ↑ IFA 2013: Ford EcoSport e nuove app via SYNC AppLink motorionline.com