dstam/matchmaking

Matchmaking数据集是一个用于训练基本约会推荐模型的入门数据集。尽管数据记录较少，但由于公开的约会应用数据集稀缺，该数据集可以作为约会或配对应用推荐模型的起点。数据集包含用户特征、用户互动、用户简介、用户寻找的理想伴侣信息以及种族信息。该数据集是通过对mstz的speeddating数据集进行转换，并使用GPT4生成用户简介和理想伴侣描述而创建的。

原始信息汇总

Matchmaking Dataset

概述

这是一个入门级数据集，旨在帮助训练基本的约会推荐模型。尽管记录不多，但由于公开可用的约会应用数据集稀缺，该数据集可以帮助启动基本的推荐模型，适用于约会或配对应用。

使用方法

python from datasets import load_dataset

dataset = load_dataset("dstam/matchmaking")

数据转换

数据集可以通过以下代码转换为用户特征和行为，形成模拟关系数据库结构： python import pandas as pd class RelationalMatchMakingDataset: def init(self, dataset_dict): self.dataset_dict = dataset_dict

def as_relational_db(self):
    main_df = self.dataset_dict["train"].to_pandas()
    actions_columns = [dated_uid, dater_uid, interests_correlation,
                       dater_liked_dated, probability_dated_wants_to_date,
                       already_met_before, dater_wants_to_date,
                       dated_wants_to_date, is_match]
    actions_df = main_df[actions_columns].copy()

    user_columns = [uid, bio, looking_for, race, is_male, age,
                    same_race_importance, same_religion_importance]
    users_df = pd.DataFrame(columns=user_columns)
    unique_users = pd.concat([main_df[dater_uid], main_df[dated_uid]]).unique()

    for uid in unique_users:
        user_data = {}
        user_data[uid] = uid
        if uid in main_df[dater_uid].values:
            dater_row = main_df[main_df[dater_uid] == uid].iloc[0]
            user_data[bio] = dater_row[dater_bio]
            user_data[looking_for] = dater_row[dater_looking_for]
            user_data[race] = dater_row[dater_race]
            user_data[is_male] = dater_row[is_dater_male]
            user_data[age] = dater_row[dater_age]
            user_data[same_race_importance] = dater_row[same_race_importance_for_dater]
            user_data[same_religion_importance] = dater_row[same_religion_importance_for_dater]
        elif uid in main_df[dated_uid].values:
            dated_row = main_df[main_df[dated_uid] == uid].iloc[0]
            user_data[bio] = dated_row[dated_bio]
            user_data[looking_for] = dated_row[dated_looking_for]
            user_data[race] = dated_row[dated_race]
            user_data[is_male] = dated_row[is_dated_male]
            user_data[age] = dated_row[dated_age]
            user_data[same_race_importance] = dated_row[same_race_importance_for_dated]
            user_data[same_religion_importance] = dated_row[same_religion_importance_for_dated]
        
        user_data_df = pd.DataFrame([user_data])
        users_df = pd.concat([users_df, user_data_df], ignore_index=True)

    relational_db = {
        "actions": actions_df,
        "users": users_df,
    }

    return relational_db

relational_db = RelationalMatchMakingDataset(dataset).as_relational_db()

数据结构

• user_feature.csv: 包含124个真实用户的各种特征。
• actions.csv: 包含1048个用户之间的交互。
• bio.csv: 包含每个用户的个人简介。
• looking_for.csv: 包含用户对“理想伴侣”的期望。
• races.csv: 将每个种族ID与种族名称关联。

来源

该数据集是mstz的speedating数据集的转换版本，经过重新结构化以更好地适应“用户/物品”框架，用于训练推荐模型。

引用

@misc{Matchmaking 1.0, title = {Matchmaking 1.0: An open-source starter dataset for training dating app and matchmaking recommendation models}, author = {dstam}, year = {2024}, publisher = {HuggingFace}, url = {https://huggingface.co/datasets/danstam/Matchmaking-1.0} }