E.Riloff & W.Lehnert; Classifying Texts Using Relevancy Signatures. In AAAI-92 Proceedings, Tenth National Conference on Artificial Chapter 9. References 58 Intellegence, AAAI Press, 1992, 329-334  Home/Search   Document Not in Database   Summary   Related Articles   Check

....such as keyword or probabilistic representations, or vector space representations [Salton McGill 1983] and Chapter 2. Mail Filtering and Interface Agents 7 Latent Semantic Indexing [Foltz Dumais 1992] The semantic approaches lie within the use of natural language [Ram 1991; Ram 1992; Riloff Lehnert 1992] One Information Retrieval system, SCISOR [Jacobs Rau 1990] makes use of a combination of top down and bottom up processing techniques in natural language analysis to process on line news feeds. Information Filters In the preliminary stages of the Oval project [Malone et al. 1987] Malone ....

....representing the interest and relevance of different concepts. This model is used to prune away concepts unlikely to be of interest from the story or message. An alternative approach was used by Riloff and Lehnert with their Relevancy Signatures Algorithm in classifying articles about terrorism [Riloff Lehnert 1992]. Other systems have been written to skim and summarise news articles, such as the FRUMP system [DeJong 1982] 2.2.1 Interface Agents Interface Agents are programs that provide assistance to a user for different tasks. Information filters can be seen as agents as they aid the user in handling ....

E.Riloff & W.Lehnert; Classifying Texts Using Relevancy Signatures. In AAAI-92 Proceedings, Tenth National Conference on Artificial Chapter 9. References 58 Intellegence, AAAI Press, 1992, 329-334

....score consistently, is undesirable when there are competing keywords that offset each other or when keywords have different meanings in different contexts. More sophisticated news systems include Riloff and Lehnert s text skimming approach to classification through the use of Relevancy Signatures (Riloff Lehnert, 1992). This approach is inspired by the skimming capabilities exhibited by humans in identifying texts relevant to a domain. In their system, input articles from the MUC 3 domain are classified as terrorist or non terrorist activities. The relevancy signature algorithm first requires training upon a ....

Riloff, E. & Lehnert, W. (1992). Classifying Texts Using Relevancy Signatures.

....not covered sufficiently in any general language dictionary or thesaurus, and we propose that they really require a more specialised source of information. An alternative approach to the acquisition of semantic information involves the use of a disambiguated corpus for training. Riloff and Lehnert [16] developed an algorithm to derive relevancy cues from training texts, which they used for information extraction. Soderland et al. 18] also developed a system to identify concepts in a text, by means of linguistic features which reliably identified the conceptual content of a phrase. Grefenstette ....

E. Riloff and W. Lehnert. Classifying texts using relevancy signatures. In Proc. of AAAI., 1992.

....are identical except for the initial interpretation of the word is . In the first parse this word is interpreted as a verb of being, while in the second it is a transitive verb. By maintaining detailed phrasal and clausal tagging, we are often able to infer a word s superficial semantic role [RILO92]. The following description of the syntax checking procedure shows how such tagging facilitates the efficient identification of parse tree nodes, as well as the subsequent firing of validation rules. 3.3. Syntax Checker The parser produces a set of parse trees that span the words in the input ....

Riloff, E. and W. Lehnert, "Classifying Texts Using Relevancy Signatures", Proceedings of the Tenth National Conference on Artificial Intelligence, San Jose, CA, pp. 329 - 334, July 1992.

....error rate falls below some preset threshold. To tag a new sentence, the most likely tags are assigned to each word and all rules are applied to each word, in turn. Unlike most corpus based 3 Evidence that incorporation of such knowledge can reduce sparse data problems can be found in Fisher and Riloff [1992]. 4 Examples of transformations are: 1) If the current word has tag a and the preceding (following) word has tag b, then change a to z; 2) If the current word has tag a and the preceding word has tag b and the following word has tag c, then change a to z. 20 approaches to part of speech ....

....the semantically driven modifications. The CIRCUS system has been used successfully to provide natural language processing capabilities for a variety of projects including summarization of wire service texts [Lehnert et al. 1993a, Lehnert et al. 1992, Lehnert et al. 1991] text classification [Riloff and Lehnert, 1992] , and the analysis of citation sentences in research papers [Lehnert et al. 1990] Its components will be described in the next three sections. 3.1.1 Syntactic Processing in CIRCUS CIRCUS s stack oriented syntactic analyzer segments incoming text into constituent phrases. In the tradition of ....

Riloff, E. and Lehnert, W. Classifying Texts Using Relevancy Signatures. In Proceedings of the Tenth National Conference on Artificial Intelligence, pages 329--334, San Jose, CA, 1992. AAAI Press / MIT Press.

....covered sufficiently in any general language dictionary or thesaurus, and we propose that they really require a more specialised source of information. An alternative approach to the acquisition of semantic information involves the use of a disambiguated corpus for training. Riloff and Lehnert [Riloff and Lehnert1992] developed an algorithm to derive relevancy cues from training texts, which they used for information extraction. Soderland et al. Soderland et al..1995] also developed a system to identify concepts in a text, by means of linguistic features which reliably identified the conceptual content of a ....

....Ananiadou1997] We propose that NLP of sublanguage texts really requires the incorporation of lexical semantic knowledge. Whilst semantic information has been extracted from dictionaries and corpora for applications such as information extraction, dictionary construction and knowledge acquisition [Riloff and Lehnert1992] Grefenstette1994] Reinert1986] it has largely been ignored for term disambiguation. A method for term sense disambiguation relies on having a suitable means of extracting terms. Our approach uses a method for multi word automatic term recognition called NC value [Frantzi and Ananiadou1997] ....

E. Riloff and W. Lehnert. 1992. Classifying texts using relevancy signatures. In Proc. of AAAI.

.... to reduce the combinatorics of parsing [7] 11] 4] partial parsing was preferred to complete parsing; finite state models of language developed consistently [1] pattern matching techniques were coupled with adequate Knowledge Bases to get generality [6] and statistics to obtain scalability [18]. The message of the last years of research is that programs that combine sound partial parsing with other strategies seem to outperform both the traditional parsing and the no parsing approaches. La Dhl International, filiale italiana del network Dhl, ha conseguito un utile lordo, prima delle ....

.... a pronoun if it is not introduced by a preposition: IF (AND (pronoun x2) NOT (prep x1) THEN x1 x2 Classification Most of the work in text classification was concerned with the use of classical Information Retrieval techniques such as statistical and keyword analysis, or pattern matching [18]. In the last few years pattern matching was enhanced with the use of knowledge bases [8] a linguistic parser was provided to improve the results of hierarchical pattern matching [6] The advantages of joint use of statistical and knowledge based techniques were pointed out in [12] In our ....

Ellen Riloff and Wendy Lehnert. Classifying texts using relevancy signatures. In Proceedings of the Tenth National Conference on Artificial Intelligence, San Jose, California, July 1992.

....terms are not covered sufficiently in any general language dictionary or thesaurus, and we propose that they really require a more specialised source of information. An alternative approach to the acquisition of semantic information involves the use of a disambiguated corpus for training. (Riloff and Lehnert, 1992) developed an algorithm to derive relevancy cues from training texts, which they used for information extraction. Soderland et al. 1995) also developed a system to identify concepts in a text, by means of linguistic features which reliably identified the conceptual content of a phrase. ....

E. Riloff and W. Lehnert. 1992. Classifying texts using relevancy signatures. In Proc. of AAAI.

.... We have demonstrated that casebase reasoning and machine learning techniques can be successfully applied to the disambiguation of relative pronouns [Cardie 1992a, 1992b, and 1992c] experiments have shown how CIRCUS can be used to support relevancy feedback algorithms for text classification [Riloff and Lehnert 1992, Riloff 1993b] experiments have been conducted with a statistical database derived from the MUC 3 corpus [Fisher and Riloff 1992] we are obtaining a better understanding of the issues associated with automated dictionary construction [Riloff and Lehnert 1993] and we have designed a portable ....

.... pronouns [Cardie 1992a, 1992b, and 1992c] experiments have shown how CIRCUS can be used to support relevancy feedback algorithms for text classification [Riloff and Lehnert 1992, Riloff 1993b] experiments have been conducted with a statistical database derived from the MUC 3 corpus [Fisher and Riloff 1992]; we are obtaining a better understanding of the issues associated with automated dictionary construction [Riloff and Lehnert 1993] and we have designed a portable and memory efficient part of speech tagger [Lehnert and McCarthy 1993] Our research activities since 1991 suggest that performance ....

Riloff, E. and Lehnert, W. (1992) "Classifying Texts Using Relevancy Signatures," Proceedings of the Tenth National Conference on Artificial Intelligence. pp. 329-334.

.... 17 Claire Cardie (the other graduate student who had worked on MUC 3 along with Ellen Riloff) produced an impressive collection of research papers in the year that followed MUC 3 [Cardie 1992a, 1992b, 1992c and 1992d] Ellen Riloff is producing equally impressive work in the year following MUC 4 [Riloff and Lehnert 1992], Riloff and Lehnert 1993] Riloff 1993a] Riloff 1993b] 18 MUC 3 output and MUC 4 output were subject to slightly different formatting guidelines, so the output templates shown here do not completely correspond with the answer key format presented earlier in Figure 2. To appear in Belief, ....

Riloff, E. and W. Lehnert. (1992). "Classifying Texts Using Relevancy Signatures", in Proceedings of the Tenth National Conference for Artificial Intelligence. San Jose, CA. pp. 329-334.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC