keyword_analysis.py

import copy
import sys
from typing import Any, List, Dict
from ast import literal_eval
from collections import Counter
from difflib import Differ
from itertools import chain

import gensim.corpora as corpora
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import spacy
from sklearn.cluster import KMeans
from spacy.matcher import Matcher

import csv_rw

"""
Example: LANGUAGE = "en"
"""
LANGUAGE = ""
PIPELINES = {
    "en": "en_core_web_sm",
    "fr": "fr_dep_news_sm"
}


def generate_corpus_file(
        data_list: list[list[str]],
        target_file_name="corpus.txt"
):
    """
    Generates a corpus text file at /corpora/target_file_name.txt

    :param data_list: List tokenized comments
    :param target_file_name: Target name for the corpus
    """
    id2word = corpora.Dictionary(data_list)

    original_stdout = sys.stdout

    with open(f"corpora/{target_file_name}", 'w', encoding="utf-8") as f:
        sys.stdout = f
        for token in sorted(list(id2word.token2id.keys())):
            # print(token.replace('_', ' '))
            print(token)
        sys.stdout = original_stdout


def compare_dataframes(
        df1: pd.DataFrame, corpus1path: str,
        df2: pd.DataFrame, corpus2path: str
) -> (list[str], list[str], dict, dict):
    """
    Compare two dataframes of comments (should be already processed by lda.py)

    :param df1: 1st dataframe read from /lda results (_)/
    :param corpus1path: Path to the 1st corpus
    :param df2: 2nd dataframe read from /lda results (_)/
    :param corpus2path: Path to the 2nd corpus
    :return: List of removed keywords, list of added keywords, dict of trending keywords & dict of keywords that are fading away
    """
    data_list1, data_list2 = get_tokens(df1), get_tokens(df2)
    keyword_list1, keyword_list2 = get_keywords(df1), get_keywords(df2)

    #                               removed_keywords, added_keywords
    # compare 2 corpora line-by-line
    added_keywords, removed_keywords = [], []
    with open(corpus1path, 'r', encoding="utf-8") as file1, open(corpus2path, 'r', encoding="utf-8") as file2:
        differ = Differ()
        # is this needed?
        for line in differ.compare(file1.readlines(), file2.readlines()):
            if line[0] == '+':
                added_keywords.append(line[2:-1])
            elif line[0] == '-':
                removed_keywords.append(line[2:-1])

    for keyword in copy.deepcopy(removed_keywords):
        if keyword in added_keywords:
            removed_keywords.remove(keyword)
            added_keywords.remove(keyword)

    data_list1_counted = Counter(list(chain.from_iterable(data_list1)))
    data_list2_counted = Counter(list(chain.from_iterable(data_list2)))
    # sort keywords by popularity
    removed_keywords = sorted(removed_keywords, key=data_list1_counted.get, reverse=True)
    added_keywords = sorted(added_keywords, key=data_list2_counted.get, reverse=True)
    #

    #                               trending_keywords, fading_away_keywords
    # get the difference between two dictionaries
    data_list2_counted.subtract(data_list1_counted)
    difference = dict(data_list2_counted.most_common())  # .most_common() keeps the sorting

    # remove topic keywords from the list
    topic_keywords = set(chain.from_iterable(keyword_list1)) | set(chain.from_iterable(keyword_list2))
    difference = {k: v for (k, v) in difference.items() if k not in topic_keywords}

    # use k-means to divide the dictionary of keyword popularity difference into 3 clusters
    y_pred = KMeans(n_clusters=3).fit_predict(np.asarray(list(difference.values())).reshape(-1, 1))
    plt.scatter(np.zeros(len(difference)), difference.values(), c=y_pred, marker=".", linewidths=0.1)
    plt.xticks([])
    plt.show()

    # assign variables to the clusters
    trending_keywords, fading_away_keywords = dict(), dict()
    for i in range(1, len(y_pred)):
        if y_pred[i] != y_pred[i - 1]:
            trending_keywords = dict(list(difference.items())[0:i])
            break
    for i in range(len(y_pred) - 2, -1, -1):
        if y_pred[i] != y_pred[i + 1]:
            fading_away_keywords = dict(list(difference.items())[-1:i:-1])
            break
    #
    return removed_keywords, added_keywords, trending_keywords, fading_away_keywords


def get_tokens(
        dataframe: pd.DataFrame
) -> list[list[str]]:
    """
    Read tokens from a dataframe of comments (should be already processed by lda.py)

    :param dataframe: DataFrame read from /lda results (_)/
    :return: List of lists of tokens
    """
    token_list = list(map(lambda x: literal_eval(x), dataframe["Tokens"].values.tolist()))
    return token_list


def get_keywords(
        dataframe: pd.DataFrame
) -> list[list[str]]:
    """
    Read keywords from a dataframe of comments (should be already processed by lda.py)

    :param dataframe: DataFrame read from /lda results (_)/
    :return: List of lists of keywords
    """
    keyword_list = list(map(lambda x: literal_eval(x), dataframe["Keywords"].values.tolist()))
    return keyword_list


def get_contribution(
        dataframe: pd.DataFrame
) -> list[float]:
    """
    Read contribution values from a dataframe of comments (should be already processed by lda.py)

    :param dataframe: DataFrame read from /lda results (_)/
    :return: List of theme contributions (float-valued)
    """
    contribution_list = dataframe["Contribution"].values.tolist()
    return contribution_list


def search(
        target_word: str,
        original_df: pd.DataFrame,
        data_list: list[list[str]],
        keyword_list: list[list[str]],
        contribution_list: list[float]
) -> pd.DataFrame:
    """
    Search for a word, a part of a word or a logical expression (using and, or, and, ( & )) in the given DataFrame

    :param target_word: String to search for. A logical expression can also be used
    :param original_df: DataFrame to search in (should be already processed by lda.py)
    :param data_list: List of lists of tokens (see get_tokens())
    :param keyword_list: List of lists of keywords (see get_keywords())
    :param contribution_list: List of theme contributions (see get_contribution())
    :return: DataFrame with columns "Text" and "Result" (0 or 1)
    """
    target_df = pd.DataFrame()
    target_df["Text"] = original_df["Text"]

    def input2expr(input_str: str):
        input_str = input_str.replace("(", "( ")
        input_str = input_str.replace(")", " )")
        input_str = ' '.join(["'" + word + "' in ֍" if word not in ('not', 'or', 'and', '(', ')')
                              else word
                              for word in input_str.split()])
        input_str = input_str.replace("( ", "(")
        input_str = input_str.replace(" )", ")")
        return input_str

    target_word = input2expr(target_word)
    target_df["Result"] = pd.Series
    for i in range(len(target_df)):
        if eval(target_word.replace('֍', str(data_list[i]))):
            target_df["Result"][i] = 1
        elif eval(target_word.replace('֍', str(keyword_list[i]))):
            target_df["Result"][i] = contribution_list[i]
        else:
            target_df["Result"][i] = 0

    return target_df


def rule_search(
        rule: List[List[Dict[str, Any]]],
        original_df: pd.DataFrame
) -> pd.DataFrame:
    """
    Search using a Matcher rule (see https://spacy.io/usage/rule-based-matching#adding-patterns-attributes)

    :param rule: SpaCy Matcher rule
    :param original_df: DataFrame to search in (should be already processed by lda.py)
    :return: DataFrame with columns "Text" and "Result" (0 or 1)
    """
    target_df = pd.DataFrame()
    target_df["Text"] = original_df["Text"]

    nlp = spacy.load(PIPELINES.get(LANGUAGE, "en_core_web_sm"))
    matcher = Matcher(nlp.vocab)
    matcher.add("rule", rule)

    target_df["Result"] = pd.Series
    for i, text in enumerate(target_df["Text"].values):
        doc = nlp(text)
        target_df["Result"][i] = int(bool(matcher(doc)))

    return target_df


"""
Example: 

if __name__ == "__main__":
    results = "lda results (en)/table_no_2022-08-17_1512.tsv"
    df = csv_rw.read(results)
    results1, results2 = "lda results (fr)/mini_table_no_2022-08-15_1307(dec2019).tsv", \
                         "lda results (fr)/mini_table_no_2022-08-15_1338(jan2020).tsv"
    old_df, new_df = csv_rw.read(results1), csv_rw.read(results2)
    old_data_list, new_data_list = get_tokens(old_df), get_tokens(new_df)
    generate_corpus_file(old_data_list, "old_corpus.txt")
    generate_corpus_file(new_data_list, "new_corpus.txt")
    deprecated, added, trending, fading_away = compare_dataframes(old_df, "corpora/old_corpus.txt",
                                                                  new_df, "corpora/new_corpus.txt")
    data_l = get_tokens(df)
    keyword_l = get_keywords(df)
    contribution_l = get_contribution(df)

    word4search = "(apple or onion) and not deli"
    search_res1 = search(word4search, df, data_l, keyword_l, contribution_l)

    rule4search = [[{"ORTH": "#"}, {"IS_ASCII": True, 'LIKE_NUM': False}]]
    search_res2 = rule_search(rule4search, df)
"""