Skip to content

Latest commit

 

History

History
179 lines (130 loc) · 5.79 KB

sig_terms.md

File metadata and controls

179 lines (130 loc) · 5.79 KB
Raw

Signifcant terms sentiment

with sample of comment data from 6/7-7/4 (~113K comments) using comments tagged by regex patterns, stored in analysis.titleii

Run signifcant terms query with tagged comments:

queries

significant in comments regex-tagged positive vs negative

["neutrality", "net", "not", "do", "be", "protect", "by", "isps", "it", "keep"]

GET _search
{
  "size": 0,
  "query": {
    "term": {
      "analysis.titleii": true
    }
  },
  "aggregations": {
    "significantTerms": {
      "significant_terms": {
        "field": "text_data",
        "mutual_information": {
          "include_negatives": false,
          "background_is_superset": false
        },
        "background_filter": {
          "term": {
            "analysis.titleii": false
          }
        }
      }
    }
  }
}

significant in comments regex-tagged negative vs positive

["bipartisan", "consensus", "obstructing", "smothering", "overreach", "bureaucratic", "obama's", "damaging", "unprecedented", "restore"]

significant in comments regex-tagged positive vs untagged

["trusted", "tell", "neutrality", "job", "net", "stand", "despise", "escape", "cannot", "action"]

GET _search
{
  "size": 0,
  "query": {
    "term": {
      "analysis.titleii": true
    }
  },
  "aggregations": {
    "significantTerms": {
      "significant_terms": {
        "field": "text_data",
        "mutual_information": {
          "include_negatives": false,
          "background_is_superset": false
        },
        "background_filter": {
          "bool": {
            "must_not": {
              "exists": {
                "field": "analysis.titleii"
              }
            }
          }
        }
      }
    }
  }
}

significant in comments regex-tagged negative vs untagged?

["obstructing", "smothering", "obama's", "unprecedented", "damaging", "imposed", "light", "20", "grab", "positive"]

significant in comments regex-tagged positive vs all?

["neutrality", "net", "trusted", "tell", "stand", "keep", "cannot", "current", "place", "users"]

GET _search
{
  "size": 0,
  "query": {
    "term": {
      "analysis.titleii": true
    }
  },
  "aggregations": {
    "significantTerms": {
      "significant_terms": {
        "field": "text_data",
        "mutual_information": {
          "include_negatives": false
        }
      }
    }
  }
}

significant in comments regex-tagged negative vs all?

["obstructing", "smothering", "obama's", "unprecedented", "damaging", "light", "imposed", "20", "grab", "positive"]

all positive terms

['action', 'be', 'by', 'cannot', 'cannot', 'current', 'despise', 'do', 'escape', 'isps', 'it', 'job', 'keep', 'keep', 'net', 'net', 'net', 'neutrality', 'neutrality', 'neutrality', 'not', 'place', 'protect', 'stand', 'stand', 'tell', 'tell', 'trusted', 'trusted', 'users']

popular: keep, "net neutrality", stand, tell, trusted

all negative

['20', '20', 'bipartisan', 'bureaucratic', 'consensus', 'damaging', 'damaging', 'damaging', 'grab', 'grab', 'imposed', 'imposed', 'light', 'light', "obama's", "obama's", "obama's", 'obstructing', 'obstructing', 'obstructing', 'overreach', 'positive', 'positive', 'restore', 'smothering', 'smothering', 'smothering', 'unprecedented', 'unprecedented', 'unprecedented']

popular: 20, damaging, grab, imposed, light, obama, obstructing, positive, smothering, unprecedented

testing

Run test_sig_terms.py and inspect results:

python test_sig_terms.py --http://127.0.0.1:9200/ --order desc --limit 1000

  • get 1000 best matches
  • write document id, text, and score to test_desc.csv

python test_sig_terms.py --http://127.0.0.1:9200/ --order asc --limit 1000

  • get 1000 worst matches
  • write document id, text, and score to test_asc.csv

misses

matched by query, but should be negative

About 300k into sig terms relevance, 62k identical comments, tagged bot.american-jobs

The free-market Internet was an incredible engine of economic growth, innovation, and job creation since the 1990s and has already been substantially slowed by the 2015 Net Neutrality rules. The slowdown in investment is destroying jobs and risks a big future tax hike to make up for lost private investment. Save American jobs by repealing Net Neutrality

10603304789399 460k in, tagged bot.best_cop

The Title II order created a gaping gap in privacy protections by taking the best cop, the FTC, off the beat. That is reason enough to support Chairman Pai's proposal to restore Internet freedom. Restore privacy by repealing Net Neutrality.

1060351723564, tagged bot.economics_free_zone

The FCC's Net Neutrality rules were written in the Obama White House by political staff and Tech Industry special interests who overruled the FCC's own experts. The FCC's own chief economist Tim Brennan called the rules "an economics-free zone." They should be repealed.

tagged bot.corrupt_result

Obama's Net Neutrality order was the corrupt result of a corrupt process controlled by Silicon Valley special interests. It gives some of the biggest companies in the world a free ride at the expense of consumers and should be immediately repealed!

tagging

export ES_ENDPONT="http://localhost:9200/"
cd server/fcc_analysis

review 1%:

cd experiments
python print.py --limit 5000 --sample 1 | more
cd ..

set analysis.sentiment_sig_terms_ordered to true for top matches of query for significant terms

fcc tag_sigterms --limit 5000

repeat