Entities

Entities near negative expressions in opinion articles

This is a very experimental entity recognition system. It does yield some good results when applied to sentences with direct sentence structure but bad results when that is not the case.

Method:

  1. Split opinion articles into sentences and tag them
  2. Iterate words in senteces, keeping track of proper nouns
  3. Use Afinn-corpus to get negativity score for words
  4. When getting a negative score, add previous proper noun to entities list
  5. Reset values on sentence break. (note exception below)

Often the entity is first introduced in one sentence and the negative proposition is made in the following sentence, where the actor is referencs it pronoun. Simplified example: "This has also been brought forward by the Democrats. They are stupid in their opposition of...” In these cases, the proper noun is fetched from the previous sentence.

Negative words are only counted when they are nouns, verbs or adjectives. In the case of verbs or adjectives, the reasoning is pretty solid. But when they are nouns, previous proper noun only likely to match if there is a verb in between. These restrictions are taken into account.

Rank Token Frequency
1 Trump 9
2 Greta Thunberg 5
3 Republicans 4
4 House 4
5 President Trump 4
6 Senate 3
7 Democrats 3
8 D.C 1
9 Democratic Party 1
10 Congresswoman Rodgers 1
11 Americans 1
12 Conservation Coalition 1
13 Joe Biden 1
14 President Bill Clinton 1
15 Richard Nixon 1
16 Barron Trump 1
17 Pamela Karlan 1
18 Where 1
19 Mr. Collins 1
20 President Clinton 1
21 Dan Gainor 1
22 Cats 1
23 Memory. 1
24 MSNBC 1
25 Twitter 1
26 Hitler 1
27 Jews 1
28 New York Times 1
29 Times 1
30 ASSAILING 1
31 Baquet 1
32 Turkey 1
33 U.S. 1
34 America 1
35 ABC 1
36 Sandmann 1
37 Covington 1
38 Avenatti 1
39 Stelter 1
40 Glenn Greenwald 1
41 *3 years* 1
42 Planet 1
43 Parliament 1
44 Decade 1
45 COP 1
46 September 1
47 Australia 1
48 Climate 1
49 Oxfam 1
50 Africa 1
51 December 1
52 Labour Party 1
53 Labour 1
54 Jeremy Corbyn 1
55 Sweden 1
56 Remain 1