Unit 7: Prototype design

Activity 1: Discussion

Work in pairs. Discuss your answers to the questions below.

1. Why are expert systems not a highly researched area nowadays?
2. In what scenarios are rule-based expert systems more suitable than machine learning systems?
3. What are the main components of an expert system?
4. Are all those components present in your expert system design?

Activity 2: Final project: Primary purpose

Read.

Current expert witnesses in criminal cases who need to examine evidene and present their findings to court do not have access to a system that can compare datasets. The primary purpose of your expert system is to create a system that identifies the linguistic similarity among datasets (corpora). The system should compare two or more known corpora to a questioned corpus. The system will evaluate which corpus is most similar and which is most dissimilar. The dissimilar corpus is ruled out. This continues until only one corpus remains. That corpus is identified most similar, and therefore we assume written by the same author as the questionned corpus.

Activity 3: Essential and desirable features

Check these lists of features for the expert system which show what the system needs to do.

All features

The expert system will be able to:

1. count, list and order the frequency of words essential
2. count, list and order the frequency of keywords essential
3. allow user to select reference corpus essential
4. allow user to select statistical formular for keyness
5. display the first 20 words/keywords of each dataset


6. order the known datasets by number of shared words/keywords with questionned dataset
7. display the shared words/keywords in the first 20 words/keywords of each dataset essential
8. identify the most similiar dataset based on the highest number of shared top 20 words/keywords of the questionned dataset
9. identify the least similiar dataset based on the lowest number of shared top 20 words/keywords of the questionned dataset essential
10. suggest to rule out the least similiar dataset essential
11. allow user to confirm or reject ruling out
12. continue to identify the most/least similiar dataset and rule out the least for the top 40, then 60 keywords


13. tag the part-of-speech (POS) of each word essential
14. allow the user to select any word or string and display the word or string in context essential
15. show 6 words before the target and 6 after the target word for each instance of the word in each dataset
16. allow the user to search for POS patterns following the target word, e.g. absolutely + JJ essential
17. count the number of identical POS patterns to Q in each K dataset essential
18. order the known datasets by number of identical POS patterns with the questionned dataset
19. identify the most similiar dataset based on the number of identical POS patterns with questionned dataset
20. identify the least similiar dataset based on the number of identical POS patterns with questionned dataset essential
21. suggest ruling out the least similiar dataset essential
22. allow user to confirm or reject ruling out

Essential features

The expert system will be able to:

1. count, list and order the frequency of words
2. count, list and order the frequency of keywords
3. allow user to select reference corpus


4. display the shared words/keywords in the first 20 words/keywords of each dataset
5. identify the least similiar dataset based on the lowest number of shared top 20 words/keywords of the questionned dataset
6. suggest to rule out the least similiar dataset


7. tag the part-of-speech (POS) of each word
8. allow the user to select any word or string and display the word or string in context
9. allow the user to search for POS patterns following the target word, e.g. absolutely + JJ
10. count the number of identical POS patterns in each dataset
11. identify the least similiar dataset based on the number of identical POS patterns with questionned dataset
12. suggest ruling out the least similiar dataset

Desirable features

The expert system will be able to:

1. allow user to select statistical formular for keyness
2. display the first 20 words/keywords of each dataset


3. order the known datasets by number of shared words/keywords with questionned dataset
4. identify the most similiar dataset based on the highest number of shared top 20 words/keywords of the questionned dataset
5. allow user to confirm or reject ruling out
6. continue to identify the most/least similiar dataset and rule out the least for the top 40, then 60 keywords


7. show 6 words before the target and 6 after the target word for each instance of the word in each dataset
8. order the known datasets by number of identical POS patterns with the questionned dataset
9. identify the most similiar dataset based on the number of identical POS patterns with questionned dataset
10. allow user to confirm or reject ruling out

Activity 4: Part-of-speech tags

Read.

POS tagging is the act of labelling words with a particular part of speech. The common parts of speech are noun, verb, adverb and adjective. However, most POS taggers use a much large set of tags. The most popular POS tagset has 36 tags. NLP pipelines that aim to map syntax or disambiguate meanings often use this layer. The Penn treebank tagset is shown in the table below.

Activity 5: Learning from peers

Check the slide deck and see if there are any ideas that you could use to improve your expert system. Slides were created by students and combined into a single slide deck.