Skip to content

Latest commit

 

History

History

api

README.md

sentimentalk: Sentiment Analysis of Talk Pages as Service

plumber-based API for tidy sentiment analysis of MediaWiki talk pages on MediaWiki-powered websites in R.

Usage

Input

The API endpoint accepts the following parameters:

  • page_name (starts with "Talk:")
  • api is used to specify the URL for the MediaWiki api.php to query
  • project and language
    • a Wikimedia project such as "wikipedia" or "wikiquote"
    • a language code such as "en" (English) or "de" (German)
  • format:
    • "condensed" (default)
    • "pretty" makes the output more human-friendly

Note: currently, classic (non-Flow) talk pages are unsupported because they are difficult to parse. As a result, if you try to supply a classic talk page, the API will yield an error message.

Output

If the talk page has been successfully parsed and analyzed, the JSON-formatted output will be:

  • status: "success" or "error"
  • message: helpful message in case of error
  • results:
    • An array of topics
      • For each topic, an array of shuffled posts (1st post in output is not necessarily the 1st post within the topic)
        • For each post, the following fields:
          • post: identifier of the post
          • participant: salted hash of the username (for anonymity)
          • total non-stopwords: number of words within the post that were not stop words such as "say", "none", "than", "haven't", "sees"
          • sentiment expression: proportion of words that express a particular sentiment / total non-stopwords (may sum up to greater than 1)

Setup

# install.packages("devtools")
devtools::install_github("bearloga/wmf-wmhack17/sentimentalk")

Example

To analyze Talk:Cross-wiki Search Result Improvements on MediaWiki we need to provide the API url "www.mediawiki.org/w/api.php"

R

sentimentalk::process (used by the endpoint) outputs a tidy tibble by default, which can be used to perform additional analyses in R:

sentiment_breakdown <- sentimentalk::process(
  page_name = "Talk:Cross-wiki Search Result Improvements",
  api = "www.mediawiki.org/w/api.php"
)
head(sentiment_breakdown)
topic post participant sentiment total non-stopwords instances of expression
“The Plan” needs update 1 928266f77ada92505be8189fa708f10e anticipation 10 1
“The Plan” needs update 1 928266f77ada92505be8189fa708f10e none/other 10 8
“The Plan” needs update 1 928266f77ada92505be8189fa708f10e positive 10 1
“The Plan” needs update 2 8a90a3c5c9df8aa7b01118299a301a08 none/other 1 1
Do we want these new search results to work across all Wikimedia projects? 1 446575c6def055dc79082ece2555062f anticipation 37 2
Do we want these new search results to work across all Wikimedia projects? 1 446575c6def055dc79082ece2555062f fear 37 1

Endpoint

We start the endpoint for local use via endpoint.R and pm2

GET

curl -s -G \
  --data-urlencode "page_name=Talk:Cross-wiki Search Result Improvements" \
  --data-urlencode "api=www.mediawiki.org/w/api.php" \
  "https://sentimentalk.wmflabs.org/analyze"

POST

curl -s \
  --data-urlencode "page_name=Talk:Cross-wiki Search Result Improvements" \
  --data-urlencode "api=www.mediawiki.org/w/api.php" \
  "https://sentimentalk.wmflabs.org/analyze"

POST with input data as JSON

curl -s \
  --data '{"page_name":"Talk:Cross-wiki Search Result Improvements", "api":"www.mediawiki.org/w/api.php"}' \
  "https://sentimentalk.wmflabs.org/analyze"

JSON output

Note: the output has been truncated.

{
  "status": "success",
  "message": "successfully retrieved",
  "results": {
    "topics": [
      {
        "topic": "“The Plan” needs update",
        "posts": [
          {
            "post": 1,
            "participant": "928266f77ada92505be8189fa708f10e",
            "total non-stopwords": 10,
            "sentiment expression": {
              "anticipation": 0.1,
              "none/other": 0.8,
              "positive": 0.1
            }
          },
          {
            "post": 2,
            "participant": "8a90a3c5c9df8aa7b01118299a301a08",
            "total non-stopwords": 1,
            "sentiment expression": {
              "none/other": 1
            }
          }
        ]
      },
      {
        "topic": "Do we want these new search results to work across all Wikimedia projects?",
        "posts": [
          {
            "post": 1,
            "participant": "633ab52b4cd7de31ff86740cd945492f",
            "total non-stopwords": 109,
            "sentiment expression": {
              "anger": 0.0092,
              "anticipation": 0.0275,
              "fear": 0.0092,
              "joy": 0.0092,
              "negative": 0.0275,
              "none/other": 0.8991,
              "positive": 0.0275, 
          ...

Additional Information

The sentiment analysis is performed using the National Research Council Canada (NRC) Word-Emotion Association Lexicon from Saif Mohammad and Peter Turney (link).