基于知识图谱的运动推荐

基于知识图谱的的运动推荐

一、数据集

https://www.kaggle.com/datasets/edoardoba/fitness-exercises-with-animations（下载地址）

由于只是测试 第二个数据集是自己随机构造的

1、原始数据集

身体部位 运动要用到的器材 运动动图的路径 id 运动的名称 目标

用户 运动 对其评分

2、数据预处理

所有运动的名称

运动所要用到的身体部位

运动要用到的器材

不同用户对不同运动的评分

import pandas as pd
import re
pd.set_option('display.max_columns', None)


def FITNESS_EXERCISE():
    pattern = re.compile("[A-Za-z0-9]+")
    out_sports = open("dataset2/out_sports.csv", "w", encoding='utf-8')
    out_sports.write("title\n")
    out_bodypart = open("dataset2/out_bodypart.csv","w",encoding='utf-8')           #写出来的csv存放在哪
    out_bodypart.write("title,bodypart\n")
    out_equipment = open("dataset2/out_equipment.csv","w",encoding='utf-8')
    out_equipment.write("title,equipment\n")
    out_gifUrl = open("dataset2/out_gifUrl.csv","w",encoding='utf-8')
    out_gifUrl.write("title,gifUrl\n")
    out_grade = open("dataset2/out_grade.csv", "w", encoding='utf-8')
    out_grade.write("user_id,title,grade\n")

    df = pd.read_csv("dataset2/fitness_exercises.csv", sep=",")
    df1 = pd.read_csv("dataset2/combined.csv", sep=",")
    print(df.head())

    d_sport = dict()
    s_sport = set()
    for _, row in df.iterrows():
        sport_id = row['id']
        title = row['name']
        d_sport[sport_id] = title
        if title in s_sport:
            continue
        s_sport.add(title)
        out_sports.write(f"{title}\n")

        bodypart = row['bodyPart']
        bodypart = "\"" + bodypart + "\""
        out_bodypart.write(f"{title},{bodypart}\n")

        equipment = row['equipment']
        equipment = "\"" + equipment + "\""
        out_equipment.write(f"{title},{equipment}\n")

        gifUrl = row['gifUrl']
        gifUrl = "\"" + gifUrl + "\""
        out_gifUrl.write(f"{title},{gifUrl}\n")

    for _, row in df1.iterrows():
        user_id = row['user_id']
        sport_id = row['sports_id']
        if(sport_id in d_sport):
            title = d_sport[sport_id]
        grade = row['grade']

        out_grade.write(f"{user_id},{title},{grade}\n")


    # print(d_sport)
    # print(s_sport)
    
if __name__ == "__main__":
    FITNESS_EXERCISE()

生成的csv文件如下

将这几个csv文件放到 D:\Neo4j\neo4j-community-4.4.8\import（下载的neo4j的安装目录下）

3、用neo4j 构建知识图谱

启动neo4j （neo4j需要配置环境变量）具体见https://blog.csdn.net/xuan314708889/article/details/103858493?ops_request_misc=%257B%2522request%255Fid%2522%253A%2522165922866416782388053719%2522%252C%2522scm%2522%253A%252220140713.130102334..%2522%257D&request_id=165922866416782388053719&biz_id=0&utm_medium=distribute.pc_search_result.none-task-blog-2_allsobaiduend~default-1-103858493-null-null.142^v35pc_rank_34&utm_term=neo4j%E9%85%8D%E7%BD%AE%E7%8E%AF%E5%A2%83%E5%8F%98%E9%87%8F&spm=1018.2226.3001.4187

打开cmd 输入 neo4j.bat console

在浏览器打开 http://localhost:7474/

（每个运动有属性：器材 身体部位 gifURL ）

（ 每个用户有对其做过的运动的评分 ）

4、通过计算用户与用户之间的相似度（余弦相似度）进行推荐

（基于用户的协调过滤推荐）

（计算每个用户之间的相似度）

#基于用户的协同过滤推荐
from neo4j import GraphDatabase
import pandas as pd

uri = "neo4j://localhost:7687"
driver = GraphDatabase.driver(uri, auth=("neo4j", "12345"))

k = 10 # nearest neighbors (most similar users) to consider
sports_common = 1 # how many movies in common to be consider an user similar
users_common = 1 # minimum number of similar users that have seen the movie to consider it
threshold_sim = 0.1 # threshold to consider users similar

def load_data():
    with driver.session() as session:
        session.run("""MATCH ()-[r]->() DELETE r""")
        session.run("""MATCH (r) DELETE r""")
        
        print("Loading sports...")
        #加载数据，创建Movie标签,title属性的实体
        session.run("""
            LOAD CSV WITH HEADERS FROM "file:///out_sports.csv" AS csv
            CREATE (:Sport {title: csv.title})
            """)

        print("Loading gradings...")
        #加载评分数据，    MERGE是搜索给定模式，如果存在，则返回结果如果它不存在于图中，则它创建新的节点/关系并返回结果。
        session.run("""
            LOAD CSV WITH HEADERS FROM "file:///out_grade.csv" AS csv
            MERGE (m:Sport {title: csv.title}) 
            MERGE (u:User {id: toInteger(csv.user_id)})
            CREATE (u)-[:RATED {grading : toInteger(csv.grade)}]->(m)
            """)
        #加载影片类型数据    
        print("Loading bodypart...")
            
        session.run("""
            LOAD CSV WITH HEADERS FROM "file:///out_bodypart.csv" AS csv
            MERGE (m:Sport {title: csv.title})
            MERGE (g:BodyPart {bodypart: csv.bodypart})
            CREATE (m)-[:HAS_BODYPART]->(g)
            """)
            
        print("Loading equipment...")
        #加载关键词数据    
        session.run("""
            LOAD CSV WITH HEADERS FROM "file:///out_equipment.csv" AS csv
            MERGE (m:Sport {title: csv.title})
            MERGE (k:Equipment {equipment: csv.equipment})
            CREATE (m)-[:HAS_EQUIPMENT]->(k)
            """)
            
        print("Loading gifUrl...")
        #制片人    
        session.run("""
            LOAD CSV WITH HEADERS FROM "file:///out_gifUrl.csv" AS csv
            MERGE (m:Sport {title: csv.title})
            MERGE (p:GifUrl {gifUrl: csv.gifUrl})
            CREATE (m)-[:HAS_GIFURL]->(p)
            """)


def queries():
    while True:
        userid = int(input("请输入要为哪位用户推荐锻炼，输入其ID即可: "))
        m = int(input("为该用户推荐多少个锻炼呢？ "))
        
        genres = []
        if int(input("是否需要过滤掉不需要的锻炼部位?（输入0或1）")):#过滤掉不喜欢的类型
            with driver.session() as session:
                try:
                    q = session.run(f"""MATCH (g:BodyPart) RETURN g.bodypart AS bodypart""")
                    result = []
                    for i, r in enumerate(q):
                        result.append(r["bodypart"])#找到图谱中所有的电影类型
                    df = pd.DataFrame(result, columns=["bodypart"])
                    print()
                    print(df)
                    inp = input("输入不需要的锻炼类型索引即可，例如：1 2 3  ")
                    if len(inp) != 0:
                        inp = inp.split(" ")
                        genres = [df["bodypart"].iloc[int(x)] for x in inp]
                except:
                    print("Error")
                    
        with driver.session() as session:#找到当前ID评分的电影
            q = session.run(f"""
                    MATCH (u1:User {{id : {userid}}})-[r:RATED]-(m:Sport)
                    RETURN m.title AS title, r.grading AS grade
                    ORDER BY grade DESC
                    """)
            
            print()
            print("Your ratings are the following:")
            
            result = []
            for r in q:
                result.append([r["title"], r["grade"]])
                
            if len(result) == 0:
                print("No ratings found")
            else:
                df = pd.DataFrame(result, columns=["title", "grade"])
                print()
                print(df.to_string(index=False))
            print()
            
            session.run(f"""
                MATCH (u1:User)-[s:SIMILARITY]-(u2:User)
                DELETE s
                """)
            #找到当前用户评分的电影以及这些电影被其他用户评分的用户，with是把查询集合当做结果以便后面用where 余弦相似度计算
            session.run(f"""
                MATCH (u1:User {{id : {userid}}})-[r1:RATED]-(m:Sport)-[r2:RATED]-(u2:User)
                WITH
                    u1, u2,
                    COUNT(m) AS sports_common,
                    SUM(r1.grading * r2.grading)/(SQRT(SUM(r1.grading^2)) * SQRT(SUM(r2.grading^2))) AS sim
                WHERE sports_common >= {sports_common} AND sim > {threshold_sim}
                MERGE (u1)-[s:SIMILARITY]-(u2)
                SET s.sim = sim
                """)
                
            Q_GENRE = ""
            if (len(genres) > 0):
                Q_GENRE = "AND ((SIZE(gen) > 0) AND "
                Q_GENRE += "(ANY(x IN " + str(genres) + " WHERE x IN gen))"
                Q_GENRE += ")"
            #找到相似的用户，然后看他们喜欢什么电影 Collect：将所有值收集到一个集合list中
            q = session.run(f"""
                    MATCH (u1:User {{id : {userid}}})-[s:SIMILARITY]-(u2:User)
                    WITH u1, u2, s
                    ORDER BY s.sim DESC LIMIT {k}
                    MATCH (m:Sport)-[r:RATED]-(u2)
                    OPTIONAL MATCH (g:BodyPart)--(m)
                    WITH u1, u2, s, m, r, COLLECT(DISTINCT g.bodypart) AS gen
                    WHERE NOT((m)-[:RATED]-(u1)) {Q_GENRE}
                    WITH
                        m.title AS title,
                        SUM(r.grading * s.sim)/SUM(s.sim) AS grade,
                        COUNT(u2) AS num,
                        gen
                    WHERE num >= {users_common}
                    RETURN title, grade, num, gen
                    ORDER BY grade DESC, num DESC
                    LIMIT {m}
                    """)

            print("Recommended sports:")

            result = []
            for r in q:
                result.append([r["title"], r["grade"], r["num"], r["gen"]])
            if len(result) == 0:
                print("No recommendations found")
                print()
                continue
            df = pd.DataFrame(result, columns=["title", "avg grade", "num recommenders", "bodypart"])
            print()
            print(df.to_string(index=False))
            print()


if __name__ == "__main__":
    if int(input("是否需要重新加载并创建知识图谱？（请选择输入0或1）")):
        load_data()
    queries()

运行结果如下