![Precision: it is the fraction of system sentence which were identified by it
correctly.
F-1 Score= it refers towards harmonic mean of precision and recall
F1 = 2. Precision .recall
Precision + Recall
Automated metrics ROUGE (Recall-oriented Understudy for Gisting Evaluation).
It is much efficient as compare to other methods.
References:
(1). Hovy, E. H. Automated Text Summarization. In R. Mitkov (ed), The Oxford
Handbookof Computational Linguistics, chapter 32, pages 583–598. Oxford
University Press, 2005.
(2). [39] Mani, I., House, D., Klein, G., et al. The TIPSTER SUMMAC Text
Summarization Evaluation. In Proceedings of EACL, 1999.
(3). Luhn, H., P. The Automatic Creation of Literature Abstracts. In Inderjeet Mani
and Mark Marbury, editors, Advances in Automatic Text Summarization. MIT
Press, 1999](https://image.slidesharecdn.com/computationalassignment2-170909103600/85/Summarization-in-Computational-linguistics-6-320.jpg)
Summarization in Computational linguistics
- 1. Name: Ahmad Mashhood Subject: Computational Linguistics Assignment Topic Summarization of single documents technical articles by computer summarizer tools. Submitted to: Sir. Aamir Wali FAST- NU Lahore Campus
- 2. Summarization: Hovey, E. H. (2005) defined summary as a text which is formulated on the criteria of “Significant portion of information from the original text ” and the criteria of its length in comparisonwith the original text which should not be more than the half of the original text. Mani and Maybury (1999) described the process of“TextSummarization” they considered it a process like “distillation” in which the most essential and significant information from one or more text is collected and this information is presented in an “abridge form”. Automated Summarization: Human made summaries generally done using human intelligent capabilities. The advancement in computer processingsystems and Natural language processing opened a new domain of research, whose focus was to producehuman like abstractive summaries of single or multiple document texts. The use of computer programs, online tools, resulted in “auto abstract” a term coined by Luhen (1958). He was the first personto give a significant work in the field of automated summarization. Needof automated summaries: The need of automated summaries is based on well-defined purposes and goals. With the advancement of extensive text on the computer web pages, document archives, various newspaper articles, reports, based on the same event it became difficult to read this kind of extensive information by human beings. It requires time, human intelligence resources, which results in difficulty in decision making process.If automated summaries are provided in no time constraints, it will save human resources, effort, and will facilitate decision making process. Goalof Automated Summarization research: The goal of this domain of research of Natural language processing (NLP) and artificial intelligence is to achieve the generation of automated abstracts with high human like similarity. Although more work is needed to move forward from extractive summaries.
- 3. Difference betweenabstractand extract summary: The basic difference between abstract and extract is (1)In extracts different words or sentences are selected from the original text and then they are combined together using a chronological sequence of the original text. Key words are also identified by using extracting technique. (2) While abstracts are better oriented and sequenced, in which words are paraphrased or new words are used and they have or should have the ability to replace the original text. Research is happening to achieve this level. Types of summaries: The summaries can be generally classified on the basis of extraction or abstraction but there are many new kinds of research oriented summaries are produced which are following, it can be Outline of a document text, main heading of a news articles, snippets, which are formed by giving a summary of a web page, when we searchthrough searchengine. It can be single document summary or multi document summary. Generic summaries: These summaries gave us significant information as their focus is not on the kind or user relevant information. Query based Summaries: when computer have gave answer to complex questions it uses the process ofsummarization after information retrieval and gave a user relevant information based summary. Snippets are also summaries Single documents summarization of technical articles: The summaries which are formed by single documents of technical articles are generally extracts. There are three general steps or problems as quoted by Martin and Juffrusky. Whenever a computer programme has to summarize a single document text article.
- 4. (1). which part of the original text content should be selected? The content selectionforsummarization should be at the level of sentences while summarizing single documents technical article. It is generally assumed before programming a summarizer tool. (2). the second problem is related to the arrangement of the extracted sentences. This ordering of information decides the structures of the summary. (3). the third problem is to make the arranged sentences fit into the context of the summary. Which is known as sentence realization. In order to achieve this stage we have to reject certain portion of sentences while, certain portion are considerimportant for contextual clarity. Non-significant phrases are removed, many sentences showing similar words are placed together to make a coherent summary. Flow diagram of a generic single document summarizer; (1).ContentSelection: It can be of two kinds (A). Unsupervised content selection: It is just like classifying sentences using classifier. Which labels each sentence, with a binary label (Important vs unimportant) or (extract worthy vs not extract worthy). Simplest unsupervised algorithm as devised by (Luhen, 1958) it refers towards selection of more salient or Information carrying sentences. Can be calculated by frequency method but usual now day’s salience is calculated by using weighting scheme. Tf-idf Weight (wi) = tfij multiply idfi
- 5. Supervised content selection: Classification: Position:T1, p2S1, P3S1, P4S1, P1S1, P2S2 Cue phrases:in short, in conclusion, in summary. Etc. Word significance, Sentence length, shorter one, Cohesion: lexical chains, more terms for chains is a significant sentence, Probability: P (extra worthy(s) |f1, f2, f3…..Fn). > 0.5 Alignment: Alignment algorithm such as HMMs (Jing, 2002), parallel corporacan also be used. Sentence simplification: It is also known as sentence “ compression” uses algorithm of a parser or partial parser , which uses rules of elimination purposed byZarjic et al (2007), Corney et al (2006) and Vander et al (2007 a) Remove. Appositives, attribution clauses, Abbreviations without named entities, initial adverbials. Evaluation of the Summarizers: Recall: it is the fraction of sentences chosen by human that were identified by the system correctly; Recall = | System –human select overlap| | Sentence selection by human|
- 6. Precision: it is the fraction of system sentence which were identified by it correctly. F-1 Score= it refers towards harmonic mean of precision and recall F1 = 2. Precision .recall Precision + Recall Automated metrics ROUGE (Recall-oriented Understudy for Gisting Evaluation). It is much efficient as compare to other methods. References: (1). Hovy, E. H. Automated Text Summarization. In R. Mitkov (ed), The Oxford Handbookof Computational Linguistics, chapter 32, pages 583–598. Oxford University Press, 2005. (2). [39] Mani, I., House, D., Klein, G., et al. The TIPSTER SUMMAC Text Summarization Evaluation. In Proceedings of EACL, 1999. (3). Luhn, H., P. The Automatic Creation of Literature Abstracts. In Inderjeet Mani and Mark Marbury, editors, Advances in Automatic Text Summarization. MIT Press, 1999