The PATENTSCOPE search system: CLIR February 2013 Sandrine Ammann Marketing & Communications Officer
To the PATENTSCOPE search system webinar CLIR
Agenda CLIR Definition History Search with CLIR Usefulness Golden rules Technicalities Q & A session
CLIR Cross-Lingual Information Retrieval Finds synonyms in different domains Translates those found synonyms + original query into different languages
CLIR – 12 languages available NON-ASIAN Dutch English French German Italian Portuguese Russian Spanish Swedish ASIAN Chinese Japanese Korean
History
History Lower language barriers in patent search First language tool developed in-house
CLIR: the interface
CLIR: precision vs recall Precision = the ability to retrieve the most precise results. Trying to find only precisely relevant items (high precision) = miss important items because they don't use quite the same vocabulary. Recall = the ability to retrieve as many documents as possible that match or are related to a query. Trying to find all the relevant items (high recall) = often get a lot of junk.
CLIR: precision vs recall
Example: precision
Example: recall
Example: ARM
CHIP
CLIR: supervised mode 2 modes: automatic and supervised Automatic: 1 step Supervised: 4 steps
Cross-Lingual Expansion (CLIR)
Result : the query from “container” to:
Supervised mode: 1 of 4 steps
Supervised mode : 2 of 4 steps
Supervised mode : 3 of 4 steps
Crowdsourcing "is the practice of obtaining needed services, ideas, or content by soliciting contributions from a large group of people and especially from the online community rather than from traditional employees or suppliers. […] Crowdsourcing is different from an ordinary outsourcing since it is a task or problem that is outsourced to an undefined public rather than a specific body. " source: http: //en. wikipedia. org/wiki/Crowdsourcing
Supervised mode : 4 of 4 steps
First: select languages
Second: select parameters
Stemming Process that removes common ending from words by English Porter algorithm electric¦al = electric¦ity = electric electron¦ics = electron
Third: check variants
Second: check variants
Editing
Checking: IPC
Supervised mode: results
Search examples: clothes for sport Entering “sports clothing” in the Simple search interface will return 168 results Entering “sports clothing” in the CLIR interface (in automatic mode) will return 5, 449 results Entering “sports clothing” in the CLIR interface (in supervised mode) will return 1, 023 results
Why use CLIR? A) Search full text collections simultaneously in many foreign languages B) Improve significantly the number of relevant results without increasing significantly the number of irrelevant results 485 results in English titles or abstracts for “sports clothing” 575 results obtained with CLIR searching in titles or abstracts in all languages C) Have confidence in your searches: No black box: users have access to the CLIR generated Boolean queries (albeit complex) and have the full control on them D) Have a responsive system even for complex queries
Golden rules Expansion modes Keyword very specific with only 1 meaning AUTO For any other queries, SUPERVISED is recommended Variants/synonyms Select words that you would like to appear in your search results If you have too much noise in the result list, remove generic variant
Golden rules Parameters 1. Title and abstract: unconstrained distance 2. Claims: sentence/paragraph distance 3. Description: sentence/paragraph distance Stemming recommended
Technicalities Compilation of a long list of titles in language pairs Creation of in-house extraction methodology Tool learns statistical bilingual dictionaries of titles ZH FR DE EN ES KO
Technicalities Quality of dictionaries: no human intervention The more title available, the better the coverage Chinese English French German Japanese Korean Portuguese Russian Spanish Dutch Italian Swedish
Technicalities Disambiguation: process of identifying the sense of a word in a sentence. http: //en. wikipedia. org/wiki/Disambiguation_%28 disambiguation%29 Disambiguation is applied to keywords: 1. Technical domains based on the IPC 2. Synonyms selection
Future plans Improve terminology coverage of already supported languages Add other languages: over 200’ 000 titles and abstracts with associated high quality translations in English
Slides and recording + www. wipo. int/patentscope/en/webinar/index. html
patentscope@wipo. int
mulțumesc
Скачать презентацию The PATENTSCOPE search system CLIR February 2013 Sandrine
012d7acb78a2a325aa22c228e6fa1667.ppt