TecRah

In today’s age of information explosion, there are many music platforms and tons of music resources. How to find the music that meets your taste from the sea of music has become a problem for many people. Don’t worry, Python can do its magic to help us build a personalized music recommendation system.

As we all know, the style of the music, the rhythm, the singer, and many other factors affect our music preferences. The goal of this music recommendation system is to analyze the user’s music preferences based on the user’s history of music playback records, and then recommend similar styles or singers of the same type of music for the user. For example, if the user often listens to Jay Chou’s songs, the system will recommend Jay Chou’s other songs or singers with similar styles to Jay Chou, such as JJ Lin’s works.

To build a recommendation system, we must first have data. We can get data from some public music datasets, such as Million Song Dataset Million Song Dataset, which contains various metadata of songs such as song name, singer, album, song duration, music style, and other information, as well as some of the user’s playback record data.

Assuming that we have downloaded the data and stored it in a local CSV file, such as music data csv, the following is the code to read the data and do some simple preprocessing.

import pandas as pd read music data data pd read csv music data csv view the first few lines of the data print data head check if there are any missing values in the data print data isnull sum preprocess some text type data, such as the singer’s name in a uniform format def clean artist name name return name lower strip data artist data artist apply clean artist name

Next, we want to extract features from the data that are useful for recommendation. For music data, the style of the song, the singer, the rhythmic characteristics of the song can be extracted through the audio processing library, but the simplified processing here does not involve complex audio features, etc. are important features.

We can use One Hot Encoding to process the categorized features, such as music style, singer, etc., and convert them into numerical features for subsequent calculations. The following is part of the code example for building a user profile, which simply counts the number of times the user has played different styles of music and the number of songs by different artists.

Construct the user profile dictionary user profiles Traverse each row in the data set for index row in data iterrows user id row user id artist row artist genre row genre If the user id is not in the user profile dictionary, initialize the user’s profile if user id not in user profiles user profiles user id artist counts genre counts update the count of artists played by the user if artist in user profiles user id artist counts user profiles user id artist counts artist 1 else user profiles user id artist counts artist 1 Update the count of music styles played by the user if genre in user profiles user id genre counts user profiles user id genre counts genre 1 else user profiles user id genre counts genre 1

Once we have the user profiles, we can calculate the similarity between users. Here we use cosine similarity to measure the degree of similarity in music preferences between users.

现在，我们就可以使用

recommend music

函数为指定的用户推荐音乐了。比如

为用户 user 123 推荐 5 首音乐 recommended recommend music user 123 for song artist similarity in recommended print f 推荐歌曲 song artist ，相似度 similarity

通过这个案例，我们看到了 Python 在构建个性化音乐推荐系统中的强大能力。当然，这只是一个简单的示例，实际的音乐推荐系统会更加复杂，涉及到更多的音频特征处理 更精准的推荐算法优化等，但这足以让我们开启探索 Python 数据挖掘与推荐系统构建的精彩旅程。