# This example uses the supreme court corpus to compute some simple information:
# - Which justices coordinate the most to others?
# - Which justices are coordinated to the most?
import convokit
# set up corpus
corpus = convokit.Corpus(filename=convokit.download("supreme-corpus"))
coord = convokit.Coordination(corpus)
# get set of all justices
justices = corpus.users(lambda user: user.info["is-justice"])
# get set of all users
everyone = corpus.users()
# compute coordination from each justice to everyone
print("Justices, ranked by how much they coordinate to others:")
justices_to_everyone = coord.score(justices, everyone)
for justice, score in sorted(justices_to_everyone.averages_by_user().items(),
key=lambda x: x[1], reverse=True):
print(justice.name, round(score, 5))
print()
# compute coordination from everyone to each justice
print("Justices, ranked by how much others coordinate to them:")
everyone_to_justices = coord.score(everyone, justices, focus="targets")
for justice, score in sorted(everyone_to_justices.averages_by_user().items(),
key=lambda x: x[1], reverse=True):
print(justice.name, round(score, 5))
print()
def command_line_main():
parser = argparse.ArgumentParser(description="Social features toolkit.")
parser.add_argument("filename", help="file to process")
parser.add_argument("--groups",
dest="groups",
help="file containing groups")
args = parser.parse_args()
corpus = convokit.Corpus(filename=args.filename)
### coordination
coord = convokit.Coordination(corpus)
# pairwise scores
pairwise_scores = coord.pairwise_scores(
corpus.speaking_pairs(user_names_only=True))
pairwise_scores_s = {
"'" + s + "' -> '" + t + "'": v
for (s, t), v in pairwise_scores.items()
}
_, pairwise_average_scores, _, _, _ = coord.score_report(pairwise_scores_s)
out = {}
out["pairwise"] = pairwise_scores_s
out["meta"] = {"pairwise-averages": pairwise_average_scores}
# user scores
coord_to = defaultdict(lambda: defaultdict(list))
coord_from = defaultdict(lambda: defaultdict(list))
for (speaker, target), m in pairwise_scores.items():
for cat, value in m.items():
coord_to[speaker][cat].append(value)
coord_from[target][cat].append(value)
coord_to_avg, coord_from_avg = {}, {}
for user, m in coord_to.items():
coord_to[user] = {
cat: sum(values) / len(values)
for cat, values in m.items()
}
for user, m in coord_from.items():
coord_from[user] = {
cat: sum(values) / len(values)
for cat, values in m.items()
}
user_scores = {}
for user in coord_to.keys() | coord_from.keys():
user_scores[user] = {}
if user in coord_to:
user_scores[user]["out"] = coord_to[user]
user_scores[user]["out-average"] = sum(coord_to[user].values()) / \
len(coord_to[user].values())
if user in coord_from:
user_scores[user]["in"] = coord_from[user]
user_scores[user]["in-average"] = sum(coord_from[user].values()) / \
len(coord_from[user].values())
out["user"] = user_scores
# group scores
if args.groups is not None:
groups = json.load(open(args.groups, "r"))
scores = {}
score_reports = {}
for name, (a, b) in groups.items():
scores[name] = coord.score(a, b)
score_reports[name] = coord.score_report(scores[name])
out["group"] = {name: scores[name] for name in scores}
out["meta"]["group-averages"] = {
name: {
"average-by-marker": mkr,
"aggregate-1": agg1,
"aggregate-2": agg2,
"aggregate-3": agg3
}
for name, (_, mkr, agg1, agg2, agg3) in score_reports.items()
}
json.dump(out,
open(os.path.splitext(args.filename)[0] + "-out.json", "w"),
indent=2,
sort_keys=True)
if conv.n_speakers() == 2:
conv_2pers.append(conv)
return conv_2pers
if __name__ == "__main__":
with open('pickles.p', 'rb') as f:
conversations = pickle.load(f)
corpus_speakers = create_speakers(conversations)
utterance_list = create_utterances(conversations)
# Creating a convokit corpus class
BCN_corpus = Corpus(utterances=utterance_list)
# Create a convokit coordination class.
coord = convokit.Coordination()
# Fit the underlying model (calculate LIWC scores)
coord.fit(BCN_corpus)
# Transform the BCN corpus into a coordination class
# (this can deal with the LIWC scores)
coord.transform(BCN_corpus)
# Example code to create alignment scores from the 40-49 age group to all others
# Creating the speaker sets based on age groups
_10_19 = list(
BCN_corpus.iter_speakers(lambda speaker: speaker.meta['age'] > 9 and
speaker.meta['age'] < 20))
_20_29 = list(
BCN_corpus.iter_speakers(lambda speaker: speaker.meta['age'] > 19 and
speaker.meta['age'] < 30))
_30_39 = list(