peter/copbird_aufarbeitung
1
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Adds pickling of tfid and saving relevance scores

Browse source
This commit is contained in:
procrastimax 2023-06-27 23:50:26 +02:00
parent 12cfbc5222
commit d1348391f9
3 changed files with 151705 additions and 26 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

BIN
data/tfidf_matrix.pckl Normal file
View file

Binary file not shown.

151691
data/tweet_relevance.json Normal file
View file

File diff suppressed because it is too large Load diff

40
src/search_index.py
View file

@ -1,42 +1,30 @@
import pandas as pd
import numpy as np
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
import pickle
tweet_path = "data/tweets_all_combined.csv"
tfidf_pickle_path = "data/tfidf_matrix.pckl"
relevancy_score_path = "data/tweet_relevance.json"
print("Creating TFIDF Matrix")
tweets = pd.read_csv(tweet_path)
print(tweets.head())
print(f"shape: {tweets.shape}")
print(f"keys: {tweets.keys()}")
vectorizer = TfidfVectorizer()
# TODO: we could stem or lemma the words as preprocessing, but maybe this is not needed?
model = vectorizer.fit_transform([x.lower() for x in tweets["tweet_text"]])
query = "@dima973"
query_vec = vectorizer.transform([query])
similarity = cosine_similarity(query_vec, model).flatten()
# only return stuff if there is acatually a good match for it
match_idx = np.where(similarity != 0)[0]
indices = np.argsort(-similarity[match_idx])
correct_indices = match_idx[indices]
result = tweets.iloc[correct_indices]
print("Saving TFIDF Matrix")
with open(tfidf_pickle_path, "wb") as f:
pickle.dump(model, f, protocol=5)
print("Calculating relevance score and saving new csv")
like_count_weight = 1.0
retweet_count_weight = 1.0
reply_count_weight = 1.0
quote_count_weight = 1.0
tweets["relevance_score"] = np.log(1 + (tweets["like_count"] * like_count_weight) + (tweets["retweet_count"] * retweet_count_weight) + (tweets["reply_count"] * reply_count_weight) + (tweets["quote_count"] * quote_count_weight))
# TODO: maybe come up with a more intelligent relevancy_score calculation
tweets["relevancy_score"] = np.log(1 + (tweets["like_count"] * like_count_weight) + (tweets["retweet_count"] * retweet_count_weight) + (tweets["reply_count"] * reply_count_weight) + (tweets["quote_count"] * quote_count_weight))
result = tweets.iloc[correct_indices]
overall = result["relevancy_score"] * similarity[correct_indices]
print(result.loc[overall.sort_values(ascending=False).index][["tweet_text", "like_count", "retweet_count", "reply_count", "quote_count", "relevancy_score"]].head())
# TODO: save trained TfidfVectorizer matrix as pickle
# TODO: save csv with relevancy score
print("Saving relevance_scores as csv")
with open(relevancy_score_path, "w") as f:
# we have the case that some metrics like like_count can be -1, the relevancy score therefore is NaN -> so we store it as '1.0'
tweets[["tweet_id", "relevance_score"]].to_csv(relevancy_score_path, header=True, index=False, na_rep=1.0)