Jiadong Yan
Jiaming Xu
Xinyi Jiang

Jiadong Yan

May 10th 2017

1. basic search: search the description on title, lyric, album and artist_name
2. advanced search: search on every possible filed, including title, lyric, album, artist_name, location, duration, genres, year
3. More like this: search similar tracks, based on its title, lyric and similarity of the artist
4. Sorting:

• default: sort by relevance
• sort by song hotness
• sort by danceability

0. install homebrew
1. brew tap homebrew/science
2. brew install hdf5
3. pip install Cython
4. install Tables by sudo pip install git+https://github.com/PyTables/PyTables
5. install other packages mentioned in Dependency.
6. build elasticsearch as mentioned in Build Elasticsearch
7. type redis-server in terminal
8. open another terminal window, type:

cd elasticsearch-<version>
./bin/elasticsearch

9. python query.py

• cp lib/name_syn.txt [your elasticsearch path]/config/name_syn.txt
• cp lib/cat_syn.txt [your elasticsearch path]/config/cat_syn.txt
  (It is not recommended, but if you really want to let your web application access a folder outside its deployment directory. You need to add permission in java.policy file. Details see http://stackoverflow.com/questions/10454037/java-security-accesscontrolexception-access-denied-java-io-filepermission)
• open elasticsearch server: cd elasticsearch-<version> ./bin/elasticsearch
• run python ./lib/buildElaticSearch.py
• build time: 9s
• use another terminal to run redis-server

1. We support baisc title and lyrics search for whatever you want!
2. We support many filters, like duration, artist, genre, etc!
3. You can find hotttest songs near your position!!!

• hdf5
• Cython
• Flask
• PyTables
• elasticsearch_dsl
• elasticsearch
• json
• math
• redis

• https://labrosa.ee.columbia.edu/millionsong/
• https://www.musixmatch.com/

{"1":{
"trackID": string,
"title": (song's name) string,
"year": int,
"song_hotttnesss": float,
"artistName": string,
"artistID": string,
"artist_hotttnesss": float,
"artist_location":String,
"duration": (seconds) int,
"release": (album name) string,
"similar_artists": a list of (artistID) string,
"lyrics": string,
"artist_longitude": float,
"artist_latitude": float,
"artist_location": String,
"danceability": float
},
"2":{

},
...
}

We have a test corpus sample_corpus.json. To build elasticsearch with this corpus, call build method with the path of sample corpus as parameter: build("sample_corpus.json")

1. simple query: {'description': u'love'}}
2. advanced search query: d_query = {'title': u'', 'lyric': u'', 'album': u'', 'max_longitude': u'', 'min_longitude': u'', 'description': u'love', 'max_duration': u'','min_duration': u'', 'artist_name': u'', 'min_latitude': u'', 'max_latitude': u'', 'year': u'', 'genre': u'', 'artist_location': u''}

The parameter is the track_id

use 'hot' or 'dance' as parameter, the results will sort by song_hotttnesss or danceability.

search({'description': u'love'},'hot') search(d_query,'hot') search_track(1,'dance')

• Main entry or runtime app of the Music Information Retrieval System.
• Integrate all the models and handle all the http request.
• Search Algorithm implementation.
• Session management.

css file and images

• search box view
• search results view

the corpus of 10000 songs, the format is described above.

This file defines a doc_tpye track and its field, as well as the analyzers

This file builds the elasticsearch index from the music_corpus.json

This file takes different types of query as input, builds elasticsearch search query, search in elastic and return the results. It is responsible of search on all fields and sort by specific features.

get raw information from hdf5 file ==> raw_music_corpus.json

get lyrics information according to train data.txt file ==> music_corpus.json

get lyrics information and genre information using API to MXM website ==> new_music_corpus.json

get desired attributes from the corpus