1 / 13

Final Demo of Assignment

Final Demo of Assignment. Group – 8 Maunik Shah Hemant Adil Akanksha Patel. POS Tagging using Viterbi Algorithm. Unknown Word handling: Using Previous tags Using Smoothing. POS Tagging using Viterbi Algorithm. Unknown Word handling: Using Previous tags. W n-1 |T n-1. Wn|T n.

oro
Download Presentation

Final Demo of Assignment

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Final Demo of Assignment Group – 8 Maunik Shah HemantAdil Akanksha Patel

  2. POS Tagging using Viterbi Algorithm • Unknown Word handling: • Using Previous tags • Using Smoothing

  3. POS Tagging using Viterbi Algorithm • Unknown Word handling: • Using Previous tags Wn-1|Tn-1 Wn|Tn Emission Table Transition Table

  4. POS Tagging using Viterbi Algorithm • Unknown Word handling: • Using Smoothing • P(Wn|Tn) = #(Wn|Tn) + ℷ #(T) + k ℷ • ℷ = 0.1 • K = |W| ; number of distinct words in corpus • T = tags Wn|Tn

  5. POS Tagging using Viterbi Algorithm • Accuracy by previous tag method (bigram) = 92.3702% • Accuracy by smoothing method (bigram) = 92.29% • Accuracy by smoothing method (trigram) = 92.88%

  6. Generative vs Discriminative Methods of tagging • Discriminative Method • Argmax P(tn | wn, tn-1) tn • P(tn | wn, tn-1) = #<tn,wn, tn-1> #<wn, tn-1> Accuracy = 89.07% • Generative Method Accuracy = 92.29%

  7. Next Word Prediction Algorithm • Only word model P(Wn|Wn-1) Accuracy = 12.49734 • Using POS Tags P(Wn|Tn-1,Wn-1) Accuracy = 13.16058 Perplexity :: For word model => 8703.63611651 For word-tag model => 8211.31994867

  8. POS Tagging using A* Algorithm • g =Σ - log(transition probability*Emission probability) • h = hop count * X • X=min[-log (transition Probability * Emission probability)] • Accuracy of system :: 89.87 %

  9. Parser projection ( Study assignment ) • Presentation has already given by us…

  10. NLTK ( Study assignment ) • Presentation has been prepared...

  11. YAGO • Demo has been prepared in terms of DFS.. Example output :: • Input siring 1 :: Tardeo Input siring 2 :: Egmore Tardeo Mumbai Maharastra India Tamilnadu Chennai Egmore ( Using only yagoGeoData file as knowledge Database )

  12. Conclusion of CS 626 • Natural Language Processing seems like a linguistic subject as first glance but more than that it has a lot of Machine Learning , AI algorithms and Mathematics involved.. Linguistic is base but NLP is more about involving linguistic to mathematics and AI.. • And seeing at the applications of NLP, we have to simply say that NLP is important to almost all automatic user interactive applications and that even involves future web structure.

  13. Take Outs from CS 626 • We now have a whole different approach to linguistic than we had 4 months ago.. • Before we learn this subject we could never think of even including probability with language, the two most distanced and most distasteful topics we had seen before coming here.. • If we tell others that we are learning linguistic subject here in IITB, they clueless guys may just laugh on us but we know that it has more interesting things to learn than we could learn in any other subject, that includes things about linguistics and applications of mathematics in linguistics too..

More Related