app.py

import pandas as pd
import numpy as np
import requests
from bs4 import BeautifulSoup
from googletrans import Translator
import tensorflow as tf
import gradio as gr
class AnimeRecommender:
    def __init__(self, rating_path, anime_path, synopsis_path, model_path):
        self.rating_df = pd.read_csv(rating_path)
        self.df_anime = pd.read_csv(anime_path, low_memory=True)
        self.sypnopsis_df = pd.read_csv(synopsis_path, usecols=["MAL_ID", "Name", "Genres", "sypnopsis"])
        self.model = tf.keras.models.load_model(model_path)
        self.translator = Translator()

        self._preprocess_data()

    def _preprocess_data(self):
        # User and anime ID encoding
        user_ids = self.rating_df["user_id"].unique().tolist()
        user2user_encoded = {x: i for i, x in enumerate(user_ids)}
        anime_ids = self.rating_df["anime_id"].unique().tolist()
        anime2anime_encoded = {x: i for i, x in enumerate(anime_ids)}

        self.rating_df["user"] = self.rating_df["user_id"].map(user2user_encoded)
        self.rating_df["anime"] = self.rating_df["anime_id"].map(anime2anime_encoded)

        self.n_users = len(user2user_encoded)
        self.n_animes = len(anime2anime_encoded)

        self.anime2anime_encoded = anime2anime_encoded
        self.anime_encoded2anime = {i: x for i, x in enumerate(anime_ids)}

        # Normalize anime weights
        self.anime_weights = self._extract_weights('anime_embedding')

        # Fix anime names
        self.df_anime['anime_id'] = self.df_anime['MAL_ID']
        self.df_anime["eng_version"] = self.df_anime['English name']
        self.df_anime['eng_version'] = self.df_anime.anime_id.apply(self._get_anime_name)

        self.df_anime.sort_values(by=['Score'], inplace=True, ascending=False, kind='quicksort', na_position='last')
        self.df_anime = self.df_anime[["anime_id", "eng_version", "Score", "Genres", "Episodes", "Type", "Premiered", "Members"]]

    def _extract_weights(self, name):
        weight_layer = self.model.get_layer(name)
        weights = weight_layer.get_weights()[0]
        weights = weights / np.linalg.norm(weights, axis=1).reshape((-1, 1))
        return weights

    def _get_anime_name(self, anime_id):
        try:
            name = self.df_anime[self.df_anime.anime_id == anime_id].eng_version.values[0]
            if name is np.nan:
                name = self.df_anime[self.df_anime.anime_id == anime_id].Name.values[0]
        except:
            name = 'Unknown'
        return name

    def get_anime_frame(self, anime):
        if isinstance(anime, int):
            return self.df_anime[self.df_anime.anime_id == anime]
        if isinstance(anime, str):
            return self.df_anime[self.df_anime.eng_version == anime]

    def get_sypnopsis(self, anime):
        if isinstance(anime, int):
            return self.sypnopsis_df[self.sypnopsis_df.MAL_ID == anime].sypnopsis.values[0]
        if isinstance(anime, str):
            return self.sypnopsis_df[self.sypnopsis_df.Name == anime].sypnopsis.values[0]

    def find_similar_animes_combined(self, anime_names, n=3, return_dist=False, neg=False):
        try:
            encoded_indices = []
            input_anime_ids = []
            for name in anime_names:
                index = self.get_anime_frame(name).anime_id.values[0]
                input_anime_ids.append(index)
                encoded_index = self.anime2anime_encoded.get(index)
                encoded_indices.append(encoded_index)

            combined_weights = np.mean(self.anime_weights[encoded_indices], axis=0)
            combined_weights = combined_weights / np.linalg.norm(combined_weights)

            dists = np.dot(self.anime_weights, combined_weights)
            sorted_dists = np.argsort(dists)
            n = n + len(input_anime_ids)

            if neg:
                closest = sorted_dists[:n]
            else:
                closest = sorted_dists[-n:]

            if return_dist:
                return dists, closest

            rindex = self.df_anime
            SimilarityArr = []
            for close in closest:
                decoded_id = self.anime_encoded2anime.get(close)
                if decoded_id in input_anime_ids:
                    continue
                sypnopsis = self.get_sypnopsis(decoded_id)
                anime_frame = self.get_anime_frame(decoded_id)
                anime_name = anime_frame.eng_version.values[0]
                genre = anime_frame.Genres.values[0]
                similarity = dists[close]
                SimilarityArr.append({"anime_id": decoded_id, "name": anime_name, "similarity": similarity, "genre": genre, 'sypnopsis': sypnopsis})

            Frame = pd.DataFrame(SimilarityArr).sort_values(by="similarity", ascending=False)
            return Frame.drop(index=0)
        except Exception as e:
            print('{}!, Not Found in Anime list'.format(anime_names))
            print(str(e))
            return pd.DataFrame()

    def get_anime_url(self, name):
        anime = self.df_anime[self.df_anime['eng_version'] == name]
        if not anime.empty:
            mal_id = anime['anime_id'].values[0]
            anime_name = anime['eng_version'].values[0].replace(' ', '_').replace(':', '_').replace('!', '_')
            return f"https://myanimelist.net/anime/{mal_id}/{anime_name}"
        else:
            print(f"{name}에 해당하는 애니메이션을 찾을 수 없습니다.")
            return None

    def extract_image_url(self, url):
        try:
            response = requests.get(url)
            response.raise_for_status()
        except requests.RequestException as e:
            print(f"페이지를 가져올 수 없습니다: {e}")
            return None

        soup = BeautifulSoup(response.text, 'html.parser')
        image_tag = soup.find('img', {'data-src': True})

        if image_tag:
            return image_tag['data-src']
        else:
            print("이미지를 찾을 수 없습니다.")
            return None

    def NCF_Recommendation(self, a, b, c):
        anime_list = [a, b, c]
        anime_result = self.find_similar_animes_combined(anime_list, n=3)

        result1 = anime_result.loc[3, 'name']
        result2 = anime_result.loc[2, 'name']
        result3 = anime_result.loc[1, 'name']

        explain1 = anime_result.loc[3, 'sypnopsis']
        explain2 = anime_result.loc[2, 'sypnopsis']
        explain3 = anime_result.loc[1, 'sypnopsis']

        url1 = self.get_anime_url(result1)
        url2 = self.get_anime_url(result2)
        url3 = self.get_anime_url(result3)

        image1 = self.extract_image_url(url1)
        image2 = self.extract_image_url(url2)
        image3 = self.extract_image_url(url3)

        return result1, explain1, image1, result2, explain2, image2, result3, explain3, image3

# 파일 경로 설정
rating_path = 'https://huggingface.co/spaces/OhST/Anime_RecSys/data/rating_complete.csv'
anime_path = 'https://huggingface.co/spaces/OhST/Anime_RecSys/data/anime.csv'
synopsis_path = 'https://huggingface.co/spaces/OhST/Anime_RecSys/data/anime_with_synopsis.csv'
model_path = 'https://huggingface.co/spaces/OhST/Anime_RecSys/data/anime_model.h5'


# 객체 생성
recommender = AnimeRecommender(rating_path, anime_path, synopsis_path, model_path)

with gr.Blocks() as app:
    with gr.Row():
        a = gr.Textbox(label="너의 최애 애니 첫 번째를 작성해봐!")
        b = gr.Textbox(label="너의 최애 애니 두 번째를 작성해봐!")
        c = gr.Textbox(label="너의 최애 애니 세 번째를 작성해봐!")

    with gr.Row():
        with gr.Column():
            img1 = gr.Image(label="1번째 애니 추천")
            output1 = gr.Textbox(label="1️⃣ 첫번째 애니 추천!")
            output2 = gr.Textbox(label="첫 번째 애니 설명", interactive=False)
        with gr.Column():
            img2 = gr.Image(label="2번째 애니 추천")
            output3 = gr.Textbox(label="2️⃣ 두번째 애니 추천!")
            output4 = gr.Textbox(label="두 번째 애니 설명", interactive=False)
        with gr.Column():
            img3 = gr.Image(label="3번째 애니 추천")
            output5 = gr.Textbox(label="3️⃣ 세번째 애니 추천!")
            output6 = gr.Textbox(label="세 번째 애니 설명", interactive=False)

    btn = gr.Button("추천을 받아봅시다!")

    btn.click(
        fn=recommender.NCF_Recommendation,
        inputs=[a, b, c],
        outputs=[output1, output2, img1, output3, output4, img2, output5, output6, img3]
    )

app.launch(share=True)