Skip to content

Instantly share code, notes, and snippets.

@klintan

klintan/compute_source_trust.py

Last active January 3, 2020 21:35
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save klintan/7bb82fce7e38db017586c817a31c0cb0 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save klintan/7bb82fce7e38db017586c817a31c0cb0 to your computer and use it in GitHub Desktop.
Download ZIP

Revisions

  1. klintan revised this gist Jan 3, 2020. 1 changed file with 12 additions and 12 deletions.
    24 changes: 12 additions & 12 deletions compute_source_trust.py
    View file
    Original file line number Diff line number Diff line change
    @@ -6,15 +6,15 @@ def compute_source_trust(data, sources):
    :param sources: dict all unique sources and current scores
    :return: dict of unique sources with updated scores
    '''
    for source in sources:
    # t(w) trustworthiness of website w
    # the average confidence of all facts supplied by website/source w
    t_w = sum([confidence for confidence in data[data['source'] == source]['confidence'].values]) / len(
    data[data['source'] == source].index)
    # tau(w) trustworthiness score of website w
    # as explained in the paper, 1 - t(w) is usually quite small and multiplying many of them
    # might lead to underflow. Therefore we take the logarithm of it to better model how trustworthy a source is
    tau_w = -np.log(1 - t_w)
    # update the source score to the new score
    sources[source] = tau_w
    return sources
    for source in sources:
    # t(w) trustworthiness of website w
    # the average confidence of all facts supplied by website/source w
    t_w = sum([confidence for confidence in data[data['source'] == source]['confidence'].values]) / len(
    data[data['source'] == source].index)
    # tau(w) trustworthiness score of website w
    # as explained in the paper, 1 - t(w) is usually quite small and multiplying many of them
    # might lead to underflow. Therefore we take the logarithm of it to better model how trustworthy a source is
    tau_w = -np.log(1 - t_w)
    # update the source score to the new score
    sources[source] = tau_w
    return sources
  2. klintan created this gist Sep 7, 2018.
    20 changes: 20 additions & 0 deletions compute_source_trust.py
    View file
    Original file line number Diff line number Diff line change
    @@ -0,0 +1,20 @@
    def compute_source_trust(data, sources):
    '''
    Compute every source trustworthiness. The trustworthiness score is the average confidence of
    all facts supplied by source w
    :param data: Dataframe all facts for object O
    :param sources: dict all unique sources and current scores
    :return: dict of unique sources with updated scores
    '''
    for source in sources:
    # t(w) trustworthiness of website w
    # the average confidence of all facts supplied by website/source w
    t_w = sum([confidence for confidence in data[data['source'] == source]['confidence'].values]) / len(
    data[data['source'] == source].index)
    # tau(w) trustworthiness score of website w
    # as explained in the paper, 1 - t(w) is usually quite small and multiplying many of them
    # might lead to underflow. Therefore we take the logarithm of it to better model how trustworthy a source is
    tau_w = -np.log(1 - t_w)
    # update the source score to the new score
    sources[source] = tau_w
    return sources