231 lines
9.2 KiB
Python
231 lines
9.2 KiB
Python
import psycopg2
|
|
import numpy as np
|
|
from sklearn.metrics.pairwise import cosine_similarity
|
|
from dotenv import load_dotenv
|
|
import os
|
|
from openai import OpenAI , APIError
|
|
from langchain.embeddings import OpenAIEmbeddings
|
|
from vectData import get_specific_data, modify_similar_data, insert_data, preparing_articles, get_source_data, get_ready_data
|
|
import tiktoken
|
|
from scrapingsingle import num_tokens_from_string, slice_text_at_2k_tokens
|
|
import json
|
|
|
|
load_dotenv()
|
|
|
|
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
|
|
client = OpenAI()
|
|
embeddings = OpenAIEmbeddings()
|
|
|
|
print(f"Checking for similar!")
|
|
|
|
host = os.getenv("DB_HOST")
|
|
port = os.getenv("DB_PORT")
|
|
user = os.getenv("DB_USER")
|
|
password = os.getenv("DB_PASSWORD")
|
|
dbname = os.getenv("DB_NAME")
|
|
|
|
def calculate_cosine_similarity(v1, v2):
|
|
v1_normalized = v1 / np.linalg.norm(v1)
|
|
v2_normalized = v2 / np.linalg.norm(v2)
|
|
|
|
similarity = cosine_similarity([v1_normalized], [v2_normalized])[0][0]
|
|
return similarity
|
|
|
|
def parse_embedding_string(embedding_str):
|
|
if isinstance(embedding_str, str):
|
|
numbers = [float(num) for num in embedding_str[1:-1].split(',')]
|
|
return np.array(numbers)
|
|
elif isinstance(embedding_str, np.ndarray):
|
|
return embedding_str
|
|
else:
|
|
raise ValueError("Invalid type for embedding_str. Must be either str or np.ndarray.")
|
|
|
|
|
|
def get_titles_links_embeddings():
|
|
conn = psycopg2.connect(
|
|
host=host,
|
|
port=port,
|
|
user=user,
|
|
password=password,
|
|
dbname=dbname
|
|
)
|
|
cursor = conn.cursor()
|
|
cursor.execute('SELECT title, link, embedding FROM vectorsvevijesti WHERE ready = True;')
|
|
data = cursor.fetchall()
|
|
cursor.close()
|
|
|
|
titles = [row[0] for row in data]
|
|
links = [row[1] for row in data]
|
|
embeddings = [parse_embedding_string(row[2]) for row in data]
|
|
|
|
return titles, links, embeddings
|
|
|
|
def find_and_group_similar_articles(eps=0.5, min_samples=2, threshold=0.95):
|
|
try:
|
|
conn = psycopg2.connect(
|
|
host=host,
|
|
port=port,
|
|
user=user,
|
|
password=password,
|
|
dbname=dbname
|
|
)
|
|
|
|
with conn, conn.cursor() as cursor:
|
|
titles, links, embeddings = get_titles_links_embeddings()
|
|
|
|
processed_articles = set()
|
|
grouped_similar_articles = []
|
|
|
|
for i, (title1, link1, embedding1) in enumerate(zip(titles, links, embeddings)):
|
|
if (title1, link1) not in processed_articles:
|
|
processed_articles.add((title1, link1))
|
|
group = [(title1, link1)]
|
|
|
|
for j, (title2, link2, embedding2) in enumerate(zip(titles, links, embeddings)):
|
|
if i != j and (title2, link2) not in processed_articles:
|
|
similarity = calculate_cosine_similarity(embedding1, embedding2)
|
|
|
|
if similarity > threshold:
|
|
processed_articles.add((title2, link2))
|
|
group.append((title2, link2))
|
|
|
|
grouped_similar_articles.append(group)
|
|
|
|
return grouped_similar_articles
|
|
|
|
except psycopg2.Error as e:
|
|
print(f"Error: {e}")
|
|
return []
|
|
|
|
def processing_similar():
|
|
grouped_similar_articles_result = find_and_group_similar_articles()
|
|
|
|
if grouped_similar_articles_result:
|
|
|
|
for group in grouped_similar_articles_result:
|
|
articles = []
|
|
|
|
if len(group) > 1:
|
|
for article_tuple in group:
|
|
if len(article_tuple) >= 2:
|
|
title, link = article_tuple[:2]
|
|
article = [title, link]
|
|
articles.append(article)
|
|
l = len(articles)
|
|
if l == 2:
|
|
print("2")
|
|
a_one = articles[0][0]
|
|
a_two = articles[1][0]
|
|
|
|
get_one = get_specific_data(a_one)
|
|
get_two = get_specific_data(a_two)
|
|
|
|
text1 = get_one[0][1]
|
|
text2 = get_two[0][1]
|
|
link1 = get_one[0][2]
|
|
link2 = get_two[0][2]
|
|
if link1 != link2:
|
|
link = f"{link1}, {link2}"
|
|
else:
|
|
link = link1
|
|
|
|
ftoks = num_tokens_from_string(text1)
|
|
stoks = num_tokens_from_string(text2)
|
|
tokens = ftoks + stoks
|
|
|
|
similar_d = f"C: {a_one}, {a_two}"
|
|
|
|
modify_similar_data(similar_d, a_one)
|
|
preparing_articles(False, a_one)
|
|
|
|
modify_similar_data(similar_d, a_two)
|
|
preparing_articles(False, a_two)
|
|
|
|
print(tokens)
|
|
if tokens > 2000:
|
|
combined_text = f"{text1} {text2}"
|
|
combined_text = slice_text_at_2k_tokens(combined_text)
|
|
user_message = rf"Here is text {combined_text}, combined from 3 sources, filter text, and make news content, return as JSON only with single 'content' field"
|
|
else:
|
|
user_message = rf"Here are 2 texts {text1} {text2}, combine the following texts into a cohesive news remove any non-news related to both texts and provide the cleaned data as a JSON only with single 'content' field."
|
|
|
|
if l == 3:
|
|
print("3")
|
|
a_one = articles[0][0]
|
|
a_two = articles[1][0]
|
|
a_three = articles[2][0]
|
|
|
|
get_one = get_specific_data(a_one)
|
|
get_two = get_specific_data(a_two)
|
|
get_three = get_specific_data(a_three)
|
|
|
|
text1 = get_one[0][1]
|
|
text2 = get_two[0][1]
|
|
text3 = get_three[0][1]
|
|
link1 = get_one[0][2]
|
|
link2 = get_two[0][2]
|
|
link3 = get_three[0][2]
|
|
if link1 != link2:
|
|
if link2 != link3:
|
|
link = f"{link1}, {link2}, {link3}"
|
|
else:
|
|
link = f"{link1}, {link2}"
|
|
else:
|
|
if link2 != link3:
|
|
link = f"{link1}, {link3}"
|
|
else:
|
|
link = link1
|
|
ftoks = num_tokens_from_string(text1)
|
|
stoks = num_tokens_from_string(text2)
|
|
ttoks = num_tokens_from_string(text3)
|
|
tokens = ftoks + stoks + ttoks
|
|
|
|
similar_d = f"C: {a_one}, {a_two}, {a_three}"
|
|
modify_similar_data(similar_d, a_one)
|
|
preparing_articles(False, a_one)
|
|
|
|
modify_similar_data(similar_d, a_two)
|
|
preparing_articles(False, a_two)
|
|
|
|
modify_similar_data(similar_d, a_three)
|
|
preparing_articles(False, a_three)
|
|
|
|
print(tokens)
|
|
if tokens > 2000:
|
|
combined_text = f"{text1} {text2} {text3}"
|
|
combined_text = slice_text_at_2k_tokens(combined_text)
|
|
user_message = rf"Here is text {combined_text}, combined from 3 sources, filter text, and make news content, return as JSON only with single 'content' field"
|
|
else:
|
|
user_message = rf"Here are 3 texts {text1} {text2} and {text3}, combine the following texts into a cohesive news remove any non-news related to both texts and provide the cleaned data as a JSON only with single 'content' field."
|
|
try:
|
|
completion = client.chat.completions.create(
|
|
model="gpt-3.5-turbo",
|
|
messages=[
|
|
{"role": "system", "content": "Data analytic, Journalist and News reporter"},
|
|
{"role": "user", "content": user_message}
|
|
])
|
|
generated_text = completion.choices[0].message.content
|
|
|
|
response_data = json.loads(generated_text)
|
|
title = a_one
|
|
text = response_data["content"]
|
|
vector = embeddings.embed_query(generated_text)
|
|
|
|
insert_data(title, text, link, vector, similar_d)
|
|
print(f"Inserting combined: {title}")
|
|
|
|
except Exception as e:
|
|
print(f"Error: {e}")
|
|
print(f"Title: {a_one}")
|
|
print(f"Answer: {generated_text}")
|
|
continue
|
|
else:
|
|
print("No similar articles found.")
|
|
if __name__=="__main__":
|
|
processing_similar()
|
|
ready = get_ready_data()
|
|
if ready:
|
|
for a in ready:
|
|
print(f"Title: {a[0]}")
|
|
print(f"Link: {a[2]}")
|
|
print(f"Status: {a[3]}") |