EXP: enrich artist info data via Last.fm and produce better embeddings

This commit is contained in:
Claudio Ortolina
2026-03-01 08:22:15 +00:00
parent 5d8bf68b64
commit b35f4aed1d
8 changed files with 309 additions and 16 deletions
+17
View File
@@ -57,6 +57,23 @@ defmodule LastFm do
end
end
def get_artist_tags(musicbrainz_id, name) do
last_fm_config = last_fm_config()
case API.get_artist_tags({:musicbrainz_id, musicbrainz_id}, last_fm_config) do
{:ok, tags} ->
{:ok, tags}
{:error, :invalid_parameters} ->
# Sometimes the artist cannot be identified with the MusicBrainz ID,
# because Last.fm doesn't have that information. In that case, we try again with the artist name.
API.get_artist_tags({:name, name}, last_fm_config)
error ->
error
end
end
def get_session(token) do
last_fm_config = last_fm_config()
+42
View File
@@ -97,6 +97,20 @@ defmodule LastFm.API do
|> get_request()
end
def get_artist_tags(id_or_name_option, config) do
params =
config
|> base_params()
|> Keyword.put(:method, "artist.getTopTags")
|> put_musicbrainz_id_or_name(id_or_name_option)
config
|> new_request()
|> Req.merge(url: "/", params: params)
|> Req.Request.append_response_steps(parse_tags: &parse_artist_tags/1)
|> get_request()
end
defp put_musicbrainz_id_or_name(params, {:musicbrainz_id, musicbrainz_id}) do
Keyword.put(params, :mbid, musicbrainz_id)
end
@@ -220,6 +234,34 @@ defmodule LastFm.API do
{request, Map.put(response, :body, artists)}
end
defp parse_artist_tags({request, response}) do
tags =
case get_in(response.body, ["toptags", "tag"]) do
nil ->
[]
tags when is_list(tags) ->
tags
|> Enum.map(fn tag -> {tag["name"], parse_tag_count(tag["count"])} end)
|> Enum.filter(fn {_name, count} -> count >= 2 end)
|> Enum.sort_by(&elem(&1, 1), :desc)
|> Enum.take(15)
end
{request, Map.put(response, :body, tags)}
end
defp parse_tag_count(count) when is_integer(count), do: count
defp parse_tag_count(count) when is_binary(count) do
case Integer.parse(count) do
{n, _} -> n
:error -> 0
end
end
defp parse_tag_count(_), do: 0
defp sanitize_url(url, api_key) do
String.replace(url, api_key, "<redacted_api_key>")
end
+21
View File
@@ -191,6 +191,27 @@ defmodule MusicLibrary.Artists do
end
end
def fetch_lastfm_data(artist_id) do
artist_info = get_artist_info!(artist_id)
name = get_in(artist_info.musicbrainz_data, ["name"]) || ""
case LastFm.get_artist_tags(artist_id, name) do
{:ok, tags} ->
tag_names = Enum.map(tags, fn {tag_name, _count} -> tag_name end)
artist_info
|> ArtistInfo.changeset(%{lastfm_data: %{"tags" => tag_names}})
|> Repo.update()
{:error, _reason} ->
{:ok, artist_info}
end
end
def fetch_lastfm_data_async(artist_id) do
enqueue_worker(Worker.FetchArtistLastFmData, %{"id" => artist_id})
end
def fetch_artist_info_async(artist_id) do
enqueue_worker(Worker.FetchArtistInfo, %{"id" => artist_id})
end
+10
View File
@@ -11,6 +11,7 @@ defmodule MusicLibrary.Artists.ArtistInfo do
field :musicbrainz_data, :map, default: %{}
field :discogs_data, :map, default: %{}
field :wikipedia_data, :map, default: %{}
field :lastfm_data, :map, default: %{}
field :image_data_hash, :string
has_one :note, Note, foreign_key: :musicbrainz_id
@@ -25,6 +26,7 @@ defmodule MusicLibrary.Artists.ArtistInfo do
:musicbrainz_data,
:discogs_data,
:wikipedia_data,
:lastfm_data,
:image_data_hash
])
|> validate_required([:musicbrainz_data])
@@ -107,4 +109,12 @@ defmodule MusicLibrary.Artists.ArtistInfo do
def wikipedia_description(artist_info) do
get_in(artist_info.wikipedia_data, ["description"])
end
def lastfm_tags(artist_info) do
get_in(artist_info.lastfm_data, ["tags"]) || []
end
def lastfm_similar_artists(artist_info) do
get_in(artist_info.lastfm_data, ["similar_artists"]) || []
end
end
+121 -16
View File
@@ -8,6 +8,7 @@ defmodule MusicLibrary.Records.Similarity do
alias MusicLibrary.Artists
alias MusicLibrary.Artists.ArtistInfo
alias MusicLibrary.Notes
alias MusicLibrary.Records
alias MusicLibrary.Records.{Record, RecordEmbedding}
alias MusicLibrary.Repo
@@ -24,18 +25,23 @@ defmodule MusicLibrary.Records.Similarity do
- Genres
- Release year
- Type (album, EP, etc.)
- Artist musical style summaries (from Wikipedia, falling back to Discogs)
- Per-artist blocks: name, country, disambiguation, Wikipedia summary (500 chars),
Discogs profile excerpt (complementing Wikipedia when both available)
- Last.fm community tags and similar artists (when stored)
- User-written record notes (when present)
"""
def text_representation(%Record{} = record) do
artist_infos =
artist_infos_map =
record.artists
|> Enum.map(& &1.musicbrainz_id)
|> Artists.get_artist_infos()
|> Map.new(&{&1.id, &1})
artist_names = Record.artist_names(record)
genres = Enum.join(record.genres, ", ")
year = extract_year(record.release_date)
type = humanize_type(record.type)
note_text = record_note_text(record.musicbrainz_id)
"""
Album: #{record.title}
@@ -44,30 +50,77 @@ defmodule MusicLibrary.Records.Similarity do
Released: #{year}
Type: #{type}
#{artist_infos_summary(artist_infos)}
#{artist_blocks_summary(record.artists, artist_infos_map)}
"""
|> String.trim()
|> Kernel.<>(note_text)
end
defp artist_infos_summary([]), do: ""
defp artist_blocks_summary([], _artist_infos_map), do: ""
defp artist_infos_summary(artist_infos) do
artist_infos
|> Enum.map(&artist_info_summary/1)
defp artist_blocks_summary(artists, artist_infos_map) do
artists
|> Enum.map(fn artist ->
artist_info = Map.get(artist_infos_map, artist.musicbrainz_id)
artist_block(artist, artist_info)
end)
|> Enum.reject(&(&1 == ""))
|> Enum.join("\n\n")
end
defp artist_info_summary(artist_info) do
cond do
wikipedia_available?(artist_info) ->
wikipedia_artist_summary(artist_info)
defp artist_block(_artist, nil), do: ""
discogs_available?(artist_info) ->
discogs_artist_summary(artist_info)
defp artist_block(artist, artist_info) do
country = safe_artist_country(artist_info)
disambiguation = non_empty_string(Map.get(artist, :disambiguation))
true ->
""
header_extras = [country, disambiguation] |> Enum.reject(&is_nil/1)
header =
if header_extras == [] do
artist.name
else
"#{artist.name} (#{Enum.join(header_extras, ", ")})"
end
content = artist_content(artist_info)
if content == "" do
""
else
"#{header}:\n#{content}"
end
end
defp artist_content(artist_info) do
has_wikipedia = wikipedia_available?(artist_info)
has_discogs = discogs_available?(artist_info)
base_content =
cond do
has_wikipedia && has_discogs ->
wikipedia = wikipedia_artist_summary(artist_info)
discogs_excerpt = discogs_artist_excerpt(artist_info)
[wikipedia, discogs_excerpt] |> Enum.reject(&(&1 == "")) |> Enum.join("\n")
has_wikipedia ->
wikipedia_artist_summary(artist_info)
has_discogs ->
discogs_artist_summary(artist_info)
true ->
""
end
tags_line = lastfm_tags_line(artist_info)
similar_line = lastfm_similar_line(artist_info)
extras = [tags_line, similar_line] |> Enum.reject(&(&1 == ""))
if extras == [] do
base_content
else
[base_content | extras] |> Enum.reject(&(&1 == "")) |> Enum.join("\n")
end
end
@@ -85,7 +138,7 @@ defmodule MusicLibrary.Records.Similarity do
defp wikipedia_artist_summary(artist_info) do
description = ArtistInfo.wikipedia_description(artist_info) || ""
summary = ArtistInfo.wikipedia_summary(artist_info) || ""
truncated_summary = truncate_to_sentence(summary, 200)
truncated_summary = truncate_to_sentence(summary, 500)
[description, truncated_summary]
|> Enum.reject(&(&1 == ""))
@@ -101,6 +154,58 @@ defmodule MusicLibrary.Records.Similarity do
truncate_to_sentence(profile, 200)
end
defp discogs_artist_excerpt(artist_info) do
profile =
Map.get(artist_info.discogs_data, "profile_plaintext") ||
Map.get(artist_info.discogs_data, "profile") ||
""
truncate_to_sentence(profile, 150)
end
defp lastfm_tags_line(artist_info) do
tags = ArtistInfo.lastfm_tags(artist_info)
if tags == [] do
""
else
"Tags: #{tags |> Enum.take(10) |> Enum.join(", ")}"
end
end
defp lastfm_similar_line(artist_info) do
similar = ArtistInfo.lastfm_similar_artists(artist_info)
if similar == [] do
""
else
"Similar artists: #{similar |> Enum.take(5) |> Enum.join(", ")}"
end
end
defp safe_artist_country(artist_info) do
case artist_info.musicbrainz_data do
%{"area" => %{"name" => name}} when is_binary(name) -> name
_ -> nil
end
end
defp record_note_text(nil), do: ""
defp record_note_text(musicbrainz_id) do
case Notes.get_note(:record, musicbrainz_id) do
%{content: content} when is_binary(content) and content != "" ->
"\n\nNotes: #{content}"
_ ->
""
end
end
defp non_empty_string(nil), do: nil
defp non_empty_string(""), do: nil
defp non_empty_string(s), do: s
@doc false
def truncate_to_sentence(text, max_length) when byte_size(text) <= max_length, do: text
@@ -0,0 +1,11 @@
defmodule MusicLibrary.Worker.FetchArtistLastFmData do
use Oban.Worker, queue: :default, max_attempts: 3
@impl Oban.Worker
def perform(%Oban.Job{args: %{"id" => artist_id}}) do
case MusicLibrary.Artists.fetch_lastfm_data(artist_id) do
{:ok, _artist_info} -> :ok
error -> error
end
end
end
@@ -0,0 +1,9 @@
defmodule MusicLibrary.Repo.Migrations.AddLastfmDataToArtistInfos do
use Ecto.Migration
def change do
alter table(:artist_infos) do
add :lastfm_data, :map, default: %{}
end
end
end
@@ -0,0 +1,78 @@
defmodule MusicLibrary.Worker.FetchArtistLastFmDataTest do
use MusicLibrary.DataCase
alias MusicLibrary.Artists
alias MusicLibrary.Artists.ArtistInfo
alias MusicLibrary.Repo
alias MusicLibrary.Worker.FetchArtistLastFmData
setup do
artist_id = Ecto.UUID.generate()
Repo.insert!(%ArtistInfo{
id: artist_id,
musicbrainz_data: %{"name" => "Steven Wilson"}
})
%{artist_id: artist_id}
end
describe "perform/1" do
test "stores Last.fm tags in lastfm_data", %{artist_id: artist_id} do
Req.Test.stub(LastFm.API, fn conn ->
Req.Test.json(conn, %{
"toptags" => %{
"tag" => [
%{"name" => "progressive rock", "count" => 100},
%{"name" => "art rock", "count" => 80},
%{"name" => "psychedelic", "count" => 60}
]
}
})
end)
assert :ok = perform_job(FetchArtistLastFmData, %{"id" => artist_id})
artist_info = Artists.get_artist_info!(artist_id)
tags = ArtistInfo.lastfm_tags(artist_info)
assert "progressive rock" in tags
assert "art rock" in tags
assert "psychedelic" in tags
end
test "returns ok when Last.fm returns an error", %{artist_id: artist_id} do
Req.Test.stub(LastFm.API, fn conn ->
Req.Test.json(conn, %{"error" => 6, "message" => "Artist not found"})
end)
assert :ok = perform_job(FetchArtistLastFmData, %{"id" => artist_id})
artist_info = Artists.get_artist_info!(artist_id)
assert ArtistInfo.lastfm_tags(artist_info) == []
end
test "filters out tags with count below 2", %{artist_id: artist_id} do
Req.Test.stub(LastFm.API, fn conn ->
Req.Test.json(conn, %{
"toptags" => %{
"tag" => [
%{"name" => "progressive rock", "count" => 50},
%{"name" => "rare tag", "count" => 1},
%{"name" => "another rare", "count" => 0}
]
}
})
end)
assert :ok = perform_job(FetchArtistLastFmData, %{"id" => artist_id})
artist_info = Artists.get_artist_info!(artist_id)
tags = ArtistInfo.lastfm_tags(artist_info)
assert "progressive rock" in tags
refute "rare tag" in tags
refute "another rare" in tags
end
end
end