<?xml version="1.0" encoding="UTF-8" ?>
<oai_dc:dc schemaLocation="http://www.openarchives.org/OAI/2.0/oai_dc/ http://www.openarchives.org/OAI/2.0/oai_dc.xsd">
<dc:title>A deep multimodal approach for cold-start music recommendation</dc:title>
<dc:creator>Oramas, Sergio</dc:creator>
<dc:creator>Nieto Caballero, Oriol</dc:creator>
<dc:creator>Sordo, Mohamed</dc:creator>
<dc:creator>Serra, Xavier</dc:creator>
<dc:subject>Recommender systems</dc:subject>
<dc:subject>Deep learning</dc:subject>
<dc:subject>Multimodal</dc:subject>
<dc:subject>Music</dc:subject>
<dc:subject>Semantics</dc:subject>
<dc:description>Comunicació presentada al 2nd Workshop on Deep Learning for Recommender Systems (DLRS 2017), celebrat el 27 d'agost del 2017 a Como, Itàlia.</dc:description>
<dc:description>An increasing amount of digital music is being published daily. Music streaming services often ingest all available music, but this poses a challenge: how to recommend new artists for which prior knowledge is scarce? In this work we aim to address this so-called cold-start problem by combining text and audio information with user feedback data using deep network architectures. Our method is divided into three steps. First, artist embeddings are learned from biographies by combining semantics, text features, and aggregated usage data. Second, track embeddings are learned from the audio signal and available feedback data. Finally, artist and track embeddings are combined in a multimodal network. Results suggest that both splitting the recommendation problem between feature levels (i.e., artist metadata and audio track), and merging feature embeddings in a multimodal approach improve the accuracy of the recommendations.</dc:description>
<dc:description>This work was partially funded by the Spanish Ministry of Economy and Competitiveness under the Maria de Maeztu Units of Excellence Programme (MDM-2015-0502).</dc:description>
<dc:date>2017</dc:date>
<dc:type>info:eu-repo/semantics/conferenceObject</dc:type>
<dc:type>info:eu-repo/semantics/acceptedVersion</dc:type>
<dc:identifier>Oramas S, Sordo M, Nieto O, Serra X. A deep multimodal approach for cold-start music recommendation. In: DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7. DOI: 10.1145/3125486.3125492</dc:identifier>
<dc:identifier>http://hdl.handle.net/10230/33519</dc:identifier>
<dc:identifier>http://dx.doi.org/10.1145/3125486.3125492</dc:identifier>
<dc:language>eng</dc:language>
<dc:relation>DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7.</dc:relation>
<dc:rights>© 2017 Association for Computing Machinery</dc:rights>
<dc:rights>info:eu-repo/semantics/openAccess</dc:rights>
<dc:format>application/pdf</dc:format>
<dc:publisher>ACM Association for Computer Machinery</dc:publisher>
</oai_dc:dc>
<?xml version="1.0" encoding="UTF-8" ?>
<d:DIDL schemaLocation="urn:mpeg:mpeg21:2002:02-DIDL-NS http://standards.iso.org/ittf/PubliclyAvailableStandards/MPEG-21_schema_files/did/didl.xsd">
<d:DIDLInfo>
<dcterms:created schemaLocation="http://purl.org/dc/terms/ http://dublincore.org/schemas/xmls/qdc/dcterms.xsd">2017-12-18T09:25:49Z</dcterms:created>
</d:DIDLInfo>
<d:Item id="hdl_10230_33519">
<d:Descriptor>
<d:Statement mimeType="application/xml; charset=utf-8">
<dii:Identifier schemaLocation="urn:mpeg:mpeg21:2002:01-DII-NS http://standards.iso.org/ittf/PubliclyAvailableStandards/MPEG-21_schema_files/dii/dii.xsd">urn:hdl:10230/33519</dii:Identifier>
</d:Statement>
</d:Descriptor>
<d:Descriptor>
<d:Statement mimeType="application/xml; charset=utf-8">
<oai_dc:dc schemaLocation="http://www.openarchives.org/OAI/2.0/oai_dc/ http://www.openarchives.org/OAI/2.0/oai_dc.xsd">
<dc:title>A deep multimodal approach for cold-start music recommendation</dc:title>
<dc:creator>Oramas, Sergio</dc:creator>
<dc:creator>Nieto Caballero, Oriol</dc:creator>
<dc:creator>Sordo, Mohamed</dc:creator>
<dc:creator>Serra, Xavier</dc:creator>
<dc:description>Comunicació presentada al 2nd Workshop on Deep Learning for Recommender Systems (DLRS 2017), celebrat el 27 d'agost del 2017 a Como, Itàlia.</dc:description>
<dc:description>An increasing amount of digital music is being published daily. Music streaming services often ingest all available music, but this poses a challenge: how to recommend new artists for which prior knowledge is scarce? In this work we aim to address this so-called cold-start problem by combining text and audio information with user feedback data using deep network architectures. Our method is divided into three steps. First, artist embeddings are learned from biographies by combining semantics, text features, and aggregated usage data. Second, track embeddings are learned from the audio signal and available feedback data. Finally, artist and track embeddings are combined in a multimodal network. Results suggest that both splitting the recommendation problem between feature levels (i.e., artist metadata and audio track), and merging feature embeddings in a multimodal approach improve the accuracy of the recommendations.</dc:description>
<dc:date>2017-12-18T09:25:49Z</dc:date>
<dc:date>2017-12-18T09:25:49Z</dc:date>
<dc:date>2017</dc:date>
<dc:type>info:eu-repo/semantics/conferenceObject</dc:type>
<dc:identifier>Oramas S, Sordo M, Nieto O, Serra X. A deep multimodal approach for cold-start music recommendation. In: DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7. DOI: 10.1145/3125486.3125492</dc:identifier>
<dc:identifier>http://hdl.handle.net/10230/33519</dc:identifier>
<dc:identifier>http://dx.doi.org/10.1145/3125486.3125492</dc:identifier>
<dc:language>eng</dc:language>
<dc:relation>DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7.</dc:relation>
<dc:rights>info:eu-repo/semantics/openAccess</dc:rights>
<dc:rights>© 2017 Association for Computing Machinery</dc:rights>
<dc:publisher>ACM Association for Computer Machinery</dc:publisher>
</oai_dc:dc>
</d:Statement>
</d:Descriptor>
<d:Component id="10230_33519_1">
</d:Component>
</d:Item>
</d:DIDL>
<?xml version="1.0" encoding="UTF-8" ?>
<dim:dim schemaLocation="http://www.dspace.org/xmlns/dspace/dim http://www.dspace.org/schema/dim.xsd">
<dim:field element="contributor" lang="ca" mdschema="dc" qualifier="author">Oramas, Sergio</dim:field>
<dim:field element="contributor" lang="ca" mdschema="dc" qualifier="author">Nieto Caballero, Oriol</dim:field>
<dim:field element="contributor" lang="ca" mdschema="dc" qualifier="author">Sordo, Mohamed</dim:field>
<dim:field element="contributor" lang="ca" mdschema="dc" qualifier="author">Serra, Xavier</dim:field>
<dim:field element="date" mdschema="dc" qualifier="accessioned">2017-12-18T09:25:49Z</dim:field>
<dim:field element="date" mdschema="dc" qualifier="available">2017-12-18T09:25:49Z</dim:field>
<dim:field element="date" mdschema="dc" qualifier="issued">2017</dim:field>
<dim:field element="identifier" mdschema="dc" qualifier="citation">Oramas S, Sordo M, Nieto O, Serra X. A deep multimodal approach for cold-start music recommendation. In: DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7. DOI: 10.1145/3125486.3125492</dim:field>
<dim:field element="identifier" mdschema="dc" qualifier="uri">http://hdl.handle.net/10230/33519</dim:field>
<dim:field element="identifier" mdschema="dc" qualifier="doi">http://dx.doi.org/10.1145/3125486.3125492</dim:field>
<dim:field element="description" mdschema="dc">Comunicació presentada al 2nd Workshop on Deep Learning for Recommender Systems (DLRS 2017), celebrat el 27 d'agost del 2017 a Como, Itàlia.</dim:field>
<dim:field element="description" lang="en" mdschema="dc" qualifier="abstract">An increasing amount of digital music is being published daily. Music streaming services often ingest all available music, but this poses a challenge: how to recommend new artists for which prior knowledge is scarce? In this work we aim to address this so-called cold-start problem by combining text and audio information with user feedback data using deep network architectures. Our method is divided into three steps. First, artist embeddings are learned from biographies by combining semantics, text features, and aggregated usage data. Second, track embeddings are learned from the audio signal and available feedback data. Finally, artist and track embeddings are combined in a multimodal network. Results suggest that both splitting the recommendation problem between feature levels (i.e., artist metadata and audio track), and merging feature embeddings in a multimodal approach improve the accuracy of the recommendations.</dim:field>
<dim:field element="description" lang="en" mdschema="dc" qualifier="provenance">Made available in DSpace on 2017-12-18T09:25:49Z (GMT). No. of bitstreams: 1 oramas_deep.pdf: 605732 bytes, checksum: 92a6a376cdeff1cc44e9f235ce544f71 (MD5) Previous issue date: 2017</dim:field>
<dim:field element="description" mdschema="dc" qualifier="sponsorship">This work was partially funded by the Spanish Ministry of Economy and Competitiveness under the Maria de Maeztu Units of Excellence Programme (MDM-2015-0502).</dim:field>
<dim:field element="format" lang="ca" mdschema="dc" qualifier="mimetype">application/pdf</dim:field>
<dim:field element="language" mdschema="dc" qualifier="iso">eng</dim:field>
<dim:field element="publisher" lang="ca" mdschema="dc">ACM Association for Computer Machinery</dim:field>
<dim:field element="relation" mdschema="dc" qualifier="ispartof">DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7.</dim:field>
<dim:field element="rights" mdschema="dc">© 2017 Association for Computing Machinery</dim:field>
<dim:field element="rights" mdschema="dc" qualifier="accessRights">info:eu-repo/semantics/openAccess</dim:field>
<dim:field element="title" lang="ca" mdschema="dc">A deep multimodal approach for cold-start music recommendation</dim:field>
<dim:field element="type" mdschema="dc">info:eu-repo/semantics/conferenceObject</dim:field>
<dim:field element="type" mdschema="dc" qualifier="version">info:eu-repo/semantics/acceptedVersion</dim:field>
<dim:field element="subject" lang="en" mdschema="dc" qualifier="keyword">Recommender systems</dim:field>
<dim:field element="subject" lang="en" mdschema="dc" qualifier="keyword">Deep learning</dim:field>
<dim:field element="subject" lang="en" mdschema="dc" qualifier="keyword">Multimodal</dim:field>
<dim:field element="subject" lang="en" mdschema="dc" qualifier="keyword">Music</dim:field>
<dim:field element="subject" lang="en" mdschema="dc" qualifier="keyword">Semantics</dim:field>
</dim:dim>
<?xml version="1.0" encoding="UTF-8" ?>
<thesis schemaLocation="http://www.ndltd.org/standards/metadata/etdms/1.0/ http://www.ndltd.org/standards/metadata/etdms/1.0/etdms.xsd">
<title>A deep multimodal approach for cold-start music recommendation</title>
<creator>Oramas, Sergio</creator>
<creator>Nieto Caballero, Oriol</creator>
<creator>Sordo, Mohamed</creator>
<creator>Serra, Xavier</creator>
<description>Comunicació presentada al 2nd Workshop on Deep Learning for Recommender Systems (DLRS 2017), celebrat el 27 d'agost del 2017 a Como, Itàlia.</description>
<description>An increasing amount of digital music is being published daily. Music streaming services often ingest all available music, but this poses a challenge: how to recommend new artists for which prior knowledge is scarce? In this work we aim to address this so-called cold-start problem by combining text and audio information with user feedback data using deep network architectures. Our method is divided into three steps. First, artist embeddings are learned from biographies by combining semantics, text features, and aggregated usage data. Second, track embeddings are learned from the audio signal and available feedback data. Finally, artist and track embeddings are combined in a multimodal network. Results suggest that both splitting the recommendation problem between feature levels (i.e., artist metadata and audio track), and merging feature embeddings in a multimodal approach improve the accuracy of the recommendations.</description>
<date>2017-12-18</date>
<date>2017-12-18</date>
<date>2017</date>
<type>info:eu-repo/semantics/conferenceObject</type>
<identifier>Oramas S, Sordo M, Nieto O, Serra X. A deep multimodal approach for cold-start music recommendation. In: DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7. DOI: 10.1145/3125486.3125492</identifier>
<identifier>http://hdl.handle.net/10230/33519</identifier>
<identifier>http://dx.doi.org/10.1145/3125486.3125492</identifier>
<language>eng</language>
<relation>DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7.</relation>
<rights>info:eu-repo/semantics/openAccess</rights>
<rights>© 2017 Association for Computing Machinery</rights>
<publisher>ACM Association for Computer Machinery</publisher>
</thesis>
<?xml version="1.0" encoding="UTF-8" ?>
<record schemaLocation="http://www.loc.gov/MARC21/slim http://www.loc.gov/standards/marcxml/schema/MARC21slim.xsd">
<leader>00925njm 22002777a 4500</leader>
<datafield ind1=" " ind2=" " tag="042">
<subfield code="a">dc</subfield>
</datafield>
<datafield ind1=" " ind2=" " tag="720">
<subfield code="a">Oramas, Sergio</subfield>
<subfield code="e">author</subfield>
</datafield>
<datafield ind1=" " ind2=" " tag="720">
<subfield code="a">Nieto Caballero, Oriol</subfield>
<subfield code="e">author</subfield>
</datafield>
<datafield ind1=" " ind2=" " tag="720">
<subfield code="a">Sordo, Mohamed</subfield>
<subfield code="e">author</subfield>
</datafield>
<datafield ind1=" " ind2=" " tag="720">
<subfield code="a">Serra, Xavier</subfield>
<subfield code="e">author</subfield>
</datafield>
<datafield ind1=" " ind2=" " tag="260">
<subfield code="c">2017</subfield>
</datafield>
<datafield ind1=" " ind2=" " tag="520">
<subfield code="a">An increasing amount of digital music is being published daily. Music streaming services often ingest all available music, but this poses a challenge: how to recommend new artists for which prior knowledge is scarce? In this work we aim to address this so-called cold-start problem by combining text and audio information with user feedback data using deep network architectures. Our method is divided into three steps. First, artist embeddings are learned from biographies by combining semantics, text features, and aggregated usage data. Second, track embeddings are learned from the audio signal and available feedback data. Finally, artist and track embeddings are combined in a multimodal network. Results suggest that both splitting the recommendation problem between feature levels (i.e., artist metadata and audio track), and merging feature embeddings in a multimodal approach improve the accuracy of the recommendations.</subfield>
</datafield>
<datafield ind1="8" ind2=" " tag="024">
<subfield code="a">Oramas S, Sordo M, Nieto O, Serra X. A deep multimodal approach for cold-start music recommendation. In: DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7. DOI: 10.1145/3125486.3125492</subfield>
</datafield>
<datafield ind1="8" ind2=" " tag="024">
<subfield code="a">http://hdl.handle.net/10230/33519</subfield>
</datafield>
<datafield ind1="8" ind2=" " tag="024">
<subfield code="a">http://dx.doi.org/10.1145/3125486.3125492</subfield>
</datafield>
<datafield ind1="0" ind2="0" tag="245">
<subfield code="a">A deep multimodal approach for cold-start music recommendation</subfield>
</datafield>
</record>
<?xml version="1.0" encoding="UTF-8" ?>
<mets ID=" DSpace_ITEM_10230-33519" OBJID=" hdl:10230/33519" PROFILE="DSpace METS SIP Profile 1.0" TYPE="DSpace ITEM" schemaLocation="http://www.loc.gov/METS/ http://www.loc.gov/standards/mets/mets.xsd">
<metsHdr CREATEDATE="2022-08-25T03:36:07Z">
<agent ROLE="CUSTODIAN" TYPE="ORGANIZATION">
<name>Repositori digital de la UPF</name>
</agent>
</metsHdr>
<dmdSec ID="DMD_10230_33519">
<mdWrap MDTYPE="MODS">
<xmlData schemaLocation="http://www.loc.gov/mods/v3 http://www.loc.gov/standards/mods/v3/mods-3-1.xsd">
<mods:mods schemaLocation="http://www.loc.gov/mods/v3 http://www.loc.gov/standards/mods/v3/mods-3-1.xsd">
<mods:name>
<mods:role>
<mods:roleTerm type="text">author</mods:roleTerm>
</mods:role>
<mods:namePart>Oramas, Sergio</mods:namePart>
</mods:name>
<mods:name>
<mods:role>
<mods:roleTerm type="text">author</mods:roleTerm>
</mods:role>
<mods:namePart>Nieto Caballero, Oriol</mods:namePart>
</mods:name>
<mods:name>
<mods:role>
<mods:roleTerm type="text">author</mods:roleTerm>
</mods:role>
<mods:namePart>Sordo, Mohamed</mods:namePart>
</mods:name>
<mods:name>
<mods:role>
<mods:roleTerm type="text">author</mods:roleTerm>
</mods:role>
<mods:namePart>Serra, Xavier</mods:namePart>
</mods:name>
<mods:extension>
<mods:dateAccessioned encoding="iso8601">2017-12-18T09:25:49Z</mods:dateAccessioned>
</mods:extension>
<mods:extension>
<mods:dateAvailable encoding="iso8601">2017-12-18T09:25:49Z</mods:dateAvailable>
</mods:extension>
<mods:originInfo>
<mods:dateIssued encoding="iso8601">2017</mods:dateIssued>
</mods:originInfo>
<mods:identifier type="citation">Oramas S, Sordo M, Nieto O, Serra X. A deep multimodal approach for cold-start music recommendation. In: DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7. DOI: 10.1145/3125486.3125492</mods:identifier>
<mods:identifier type="uri">http://hdl.handle.net/10230/33519</mods:identifier>
<mods:identifier type="doi">http://dx.doi.org/10.1145/3125486.3125492</mods:identifier>
<mods:abstract>An increasing amount of digital music is being published daily. Music streaming services often ingest all available music, but this poses a challenge: how to recommend new artists for which prior knowledge is scarce? In this work we aim to address this so-called cold-start problem by combining text and audio information with user feedback data using deep network architectures. Our method is divided into three steps. First, artist embeddings are learned from biographies by combining semantics, text features, and aggregated usage data. Second, track embeddings are learned from the audio signal and available feedback data. Finally, artist and track embeddings are combined in a multimodal network. Results suggest that both splitting the recommendation problem between feature levels (i.e., artist metadata and audio track), and merging feature embeddings in a multimodal approach improve the accuracy of the recommendations.</mods:abstract>
<mods:language>
<mods:languageTerm authority="rfc3066">eng</mods:languageTerm>
</mods:language>
<mods:accessCondition type="useAndReproduction">© 2017 Association for Computing Machinery</mods:accessCondition>
<mods:titleInfo>
<mods:title>A deep multimodal approach for cold-start music recommendation</mods:title>
</mods:titleInfo>
<mods:genre>info:eu-repo/semantics/conferenceObject</mods:genre>
</mods:mods>
</xmlData>
</mdWrap>
</dmdSec>
<amdSec ID="FO_10230_33519_1">
<techMD ID="TECH_O_10230_33519_1">
<mdWrap MDTYPE="PREMIS">
<xmlData schemaLocation="http://www.loc.gov/standards/premis http://www.loc.gov/standards/premis/PREMIS-v1-0.xsd">
<premis:premis>
<premis:object>
<premis:objectIdentifier>
<premis:objectIdentifierType>URL</premis:objectIdentifierType>
<premis:objectIdentifierValue>http://repositori.upf.edu/bitstream/10230/33519/1/oramas_deep.pdf</premis:objectIdentifierValue>
</premis:objectIdentifier>
<premis:objectCategory>File</premis:objectCategory>
<premis:objectCharacteristics>
<premis:fixity>
<premis:messageDigestAlgorithm>MD5</premis:messageDigestAlgorithm>
<premis:messageDigest>92a6a376cdeff1cc44e9f235ce544f71</premis:messageDigest>
</premis:fixity>
<premis:size>605732</premis:size>
<premis:format>
<premis:formatDesignation>
<premis:formatName>application/pdf</premis:formatName>
</premis:formatDesignation>
</premis:format>
</premis:objectCharacteristics>
<premis:originalName>oramas_deep.pdf</premis:originalName>
</premis:object>
</premis:premis>
</xmlData>
</mdWrap>
</techMD>
</amdSec>
<amdSec ID="FT_10230_33519_2">
<techMD ID="TECH_T_10230_33519_2">
<mdWrap MDTYPE="PREMIS">
<xmlData schemaLocation="http://www.loc.gov/standards/premis http://www.loc.gov/standards/premis/PREMIS-v1-0.xsd">
<premis:premis>
<premis:object>
<premis:objectIdentifier>
<premis:objectIdentifierType>URL</premis:objectIdentifierType>
<premis:objectIdentifierValue>http://repositori.upf.edu/bitstream/10230/33519/2/oramas_deep.pdf.txt</premis:objectIdentifierValue>
</premis:objectIdentifier>
<premis:objectCategory>File</premis:objectCategory>
<premis:objectCharacteristics>
<premis:fixity>
<premis:messageDigestAlgorithm>MD5</premis:messageDigestAlgorithm>
<premis:messageDigest>35522f1cee331637b61d25da49f707c4</premis:messageDigest>
</premis:fixity>
<premis:size>32866</premis:size>
<premis:format>
<premis:formatDesignation>
<premis:formatName>text/plain</premis:formatName>
</premis:formatDesignation>
</premis:format>
</premis:objectCharacteristics>
<premis:originalName>oramas_deep.pdf.txt</premis:originalName>
</premis:object>
</premis:premis>
</xmlData>
</mdWrap>
</techMD>
</amdSec>
<fileSec>
<fileGrp USE="ORIGINAL">
<file ADMID="FO_10230_33519_1" CHECKSUM="92a6a376cdeff1cc44e9f235ce544f71" CHECKSUMTYPE="MD5" GROUPID="GROUP_BITSTREAM_10230_33519_1" ID="BITSTREAM_ORIGINAL_10230_33519_1" MIMETYPE="application/pdf" SEQ="1" SIZE="605732">
</file>
</fileGrp>
<fileGrp USE="TEXT">
<file ADMID="FT_10230_33519_2" CHECKSUM="35522f1cee331637b61d25da49f707c4" CHECKSUMTYPE="MD5" GROUPID="GROUP_BITSTREAM_10230_33519_2" ID="BITSTREAM_TEXT_10230_33519_2" MIMETYPE="text/plain" SEQ="2" SIZE="32866">
</file>
</fileGrp>
</fileSec>
<structMap LABEL="DSpace Object" TYPE="LOGICAL">
<div ADMID="DMD_10230_33519" TYPE="DSpace Object Contents">
<div TYPE="DSpace BITSTREAM">
</div>
</div>
</structMap>
</mets>
<?xml version="1.0" encoding="UTF-8" ?>
<mods:mods schemaLocation="http://www.loc.gov/mods/v3 http://www.loc.gov/standards/mods/v3/mods-3-1.xsd">
<mods:name>
<mods:namePart>Oramas, Sergio</mods:namePart>
</mods:name>
<mods:name>
<mods:namePart>Nieto Caballero, Oriol</mods:namePart>
</mods:name>
<mods:name>
<mods:namePart>Sordo, Mohamed</mods:namePart>
</mods:name>
<mods:name>
<mods:namePart>Serra, Xavier</mods:namePart>
</mods:name>
<mods:extension>
<mods:dateAvailable encoding="iso8601">2017-12-18T09:25:49Z</mods:dateAvailable>
</mods:extension>
<mods:extension>
<mods:dateAccessioned encoding="iso8601">2017-12-18T09:25:49Z</mods:dateAccessioned>
</mods:extension>
<mods:originInfo>
<mods:dateIssued encoding="iso8601">2017</mods:dateIssued>
</mods:originInfo>
<mods:identifier type="citation">Oramas S, Sordo M, Nieto O, Serra X. A deep multimodal approach for cold-start music recommendation. In: DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7. DOI: 10.1145/3125486.3125492</mods:identifier>
<mods:identifier type="uri">http://hdl.handle.net/10230/33519</mods:identifier>
<mods:identifier type="doi">http://dx.doi.org/10.1145/3125486.3125492</mods:identifier>
<mods:abstract>An increasing amount of digital music is being published daily. Music streaming services often ingest all available music, but this poses a challenge: how to recommend new artists for which prior knowledge is scarce? In this work we aim to address this so-called cold-start problem by combining text and audio information with user feedback data using deep network architectures. Our method is divided into three steps. First, artist embeddings are learned from biographies by combining semantics, text features, and aggregated usage data. Second, track embeddings are learned from the audio signal and available feedback data. Finally, artist and track embeddings are combined in a multimodal network. Results suggest that both splitting the recommendation problem between feature levels (i.e., artist metadata and audio track), and merging feature embeddings in a multimodal approach improve the accuracy of the recommendations.</mods:abstract>
<mods:language>
<mods:languageTerm>eng</mods:languageTerm>
</mods:language>
<mods:accessCondition type="useAndReproduction">info:eu-repo/semantics/openAccess</mods:accessCondition>
<mods:accessCondition type="useAndReproduction">© 2017 Association for Computing Machinery</mods:accessCondition>
<mods:titleInfo>
<mods:title>A deep multimodal approach for cold-start music recommendation</mods:title>
</mods:titleInfo>
<mods:genre>info:eu-repo/semantics/conferenceObject</mods:genre>
</mods:mods>
<?xml version="1.0" encoding="UTF-8" ?>
<atom:entry schemaLocation="http://www.w3.org/2005/Atom http://www.kbcafe.com/rss/atom.xsd.xml">
<atom:id>http://oai-repositori.upf.edu/oai/metadata/handle/10230/33519/ore.xml</atom:id>
<atom:published>2017-12-18T09:25:49Z</atom:published>
<atom:updated>2017-12-18T09:25:49Z</atom:updated>
<atom:source>
<atom:generator>Repositori digital de la UPF</atom:generator>
</atom:source>
<atom:title>A deep multimodal approach for cold-start music recommendation</atom:title>
<atom:author>
<atom:name>Oramas, Sergio</atom:name>
</atom:author>
<atom:author>
<atom:name>Nieto Caballero, Oriol</atom:name>
</atom:author>
<atom:author>
<atom:name>Sordo, Mohamed</atom:name>
</atom:author>
<atom:author>
<atom:name>Serra, Xavier</atom:name>
</atom:author>
<oreatom:triples>
<rdf:Description about="http://oai-repositori.upf.edu/oai/metadata/handle/10230/33519/ore.xml#atom">
<dcterms:modified>2017-12-18T09:25:49Z</dcterms:modified>
</rdf:Description>
<rdf:Description about="http://repositori.upf.edu/bitstream/10230/33519/3/oramas_deep.pdf.jpg">
<dcterms:description>THUMBNAIL</dcterms:description>
</rdf:Description>
<rdf:Description about="http://repositori.upf.edu/bitstream/10230/33519/2/oramas_deep.pdf.txt">
<dcterms:description>TEXT</dcterms:description>
</rdf:Description>
<rdf:Description about="http://repositori.upf.edu/bitstream/10230/33519/1/oramas_deep.pdf">
<dcterms:description>ORIGINAL</dcterms:description>
</rdf:Description>
</oreatom:triples>
</atom:entry>
<?xml version="1.0" encoding="UTF-8" ?>
<qdc:qualifieddc schemaLocation="http://purl.org/dc/elements/1.1/ http://dublincore.org/schemas/xmls/qdc/2006/01/06/dc.xsd http://purl.org/dc/terms/ http://dublincore.org/schemas/xmls/qdc/2006/01/06/dcterms.xsd http://dspace.org/qualifieddc/ http://www.ukoln.ac.uk/metadata/dcmi/xmlschema/qualifieddc.xsd">
<dc:title>A deep multimodal approach for cold-start music recommendation</dc:title>
<dc:creator>Oramas, Sergio</dc:creator>
<dc:creator>Nieto Caballero, Oriol</dc:creator>
<dc:creator>Sordo, Mohamed</dc:creator>
<dc:creator>Serra, Xavier</dc:creator>
<dcterms:abstract>An increasing amount of digital music is being published daily. Music streaming services often ingest all available music, but this poses a challenge: how to recommend new artists for which prior knowledge is scarce? In this work we aim to address this so-called cold-start problem by combining text and audio information with user feedback data using deep network architectures. Our method is divided into three steps. First, artist embeddings are learned from biographies by combining semantics, text features, and aggregated usage data. Second, track embeddings are learned from the audio signal and available feedback data. Finally, artist and track embeddings are combined in a multimodal network. Results suggest that both splitting the recommendation problem between feature levels (i.e., artist metadata and audio track), and merging feature embeddings in a multimodal approach improve the accuracy of the recommendations.</dcterms:abstract>
<dc:date>2017</dc:date>
<dc:type>info:eu-repo/semantics/conferenceObject</dc:type>
<dc:identifier>Oramas S, Sordo M, Nieto O, Serra X. A deep multimodal approach for cold-start music recommendation. In: DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7. DOI: 10.1145/3125486.3125492</dc:identifier>
<dc:identifier>http://hdl.handle.net/10230/33519</dc:identifier>
<dc:identifier>http://dx.doi.org/10.1145/3125486.3125492</dc:identifier>
<dc:language>eng</dc:language>
<dc:relation>DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7.</dc:relation>
<dc:rights>info:eu-repo/semantics/openAccess</dc:rights>
<dc:rights>© 2017 Association for Computing Machinery</dc:rights>
<dc:publisher>ACM Association for Computer Machinery</dc:publisher>
</qdc:qualifieddc>
<?xml version="1.0" encoding="UTF-8" ?>
<rdf:RDF schemaLocation="http://www.openarchives.org/OAI/2.0/rdf/ http://www.openarchives.org/OAI/2.0/rdf.xsd">
<ow:Publication about="oai:repositori.upf.edu:10230/33519">
<dc:title>A deep multimodal approach for cold-start music recommendation</dc:title>
<dc:creator>Oramas, Sergio</dc:creator>
<dc:creator>Nieto Caballero, Oriol</dc:creator>
<dc:creator>Sordo, Mohamed</dc:creator>
<dc:creator>Serra, Xavier</dc:creator>
<dc:description>Comunicació presentada al 2nd Workshop on Deep Learning for Recommender Systems (DLRS 2017), celebrat el 27 d'agost del 2017 a Como, Itàlia.</dc:description>
<dc:description>An increasing amount of digital music is being published daily. Music streaming services often ingest all available music, but this poses a challenge: how to recommend new artists for which prior knowledge is scarce? In this work we aim to address this so-called cold-start problem by combining text and audio information with user feedback data using deep network architectures. Our method is divided into three steps. First, artist embeddings are learned from biographies by combining semantics, text features, and aggregated usage data. Second, track embeddings are learned from the audio signal and available feedback data. Finally, artist and track embeddings are combined in a multimodal network. Results suggest that both splitting the recommendation problem between feature levels (i.e., artist metadata and audio track), and merging feature embeddings in a multimodal approach improve the accuracy of the recommendations.</dc:description>
<dc:date>2017</dc:date>
<dc:type>info:eu-repo/semantics/conferenceObject</dc:type>
<dc:identifier>Oramas S, Sordo M, Nieto O, Serra X. A deep multimodal approach for cold-start music recommendation. In: DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7. DOI: 10.1145/3125486.3125492</dc:identifier>
<dc:identifier>http://hdl.handle.net/10230/33519</dc:identifier>
<dc:identifier>http://dx.doi.org/10.1145/3125486.3125492</dc:identifier>
<dc:language>eng</dc:language>
<dc:relation>DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7.</dc:relation>
<dc:rights>info:eu-repo/semantics/openAccess</dc:rights>
<dc:rights>© 2017 Association for Computing Machinery</dc:rights>
<dc:publisher>ACM Association for Computer Machinery</dc:publisher>
</ow:Publication>
</rdf:RDF>
<?xml version="1.0" encoding="UTF-8" ?>
<metadata schemaLocation="http://www.lyncode.com/xoai http://www.lyncode.com/xsd/xoai.xsd">
<element name="dc">
<element name="contributor">
<element name="author">
<element name="ca">
<field name="value">Oramas, Sergio</field>
<field name="value">Nieto Caballero, Oriol</field>
<field name="value">Sordo, Mohamed</field>
<field name="value">Serra, Xavier</field>
</element>
</element>
</element>
<element name="date">
<element name="accessioned">
<element name="none">
<field name="value">2017-12-18T09:25:49Z</field>
</element>
</element>
<element name="available">
<element name="none">
<field name="value">2017-12-18T09:25:49Z</field>
</element>
</element>
<element name="issued">
<element name="none">
<field name="value">2017</field>
</element>
</element>
</element>
<element name="identifier">
<element name="citation">
<element name="none">
<field name="value">Oramas S, Sordo M, Nieto O, Serra X. A deep multimodal approach for cold-start music recommendation. In: DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7. DOI: 10.1145/3125486.3125492</field>
</element>
</element>
<element name="uri">
<element name="none">
<field name="value">http://hdl.handle.net/10230/33519</field>
</element>
</element>
<element name="doi">
<element name="none">
<field name="value">http://dx.doi.org/10.1145/3125486.3125492</field>
</element>
</element>
</element>
<element name="description">
<element name="none">
<field name="value">Comunicació presentada al 2nd Workshop on Deep Learning for Recommender Systems (DLRS 2017), celebrat el 27 d'agost del 2017 a Como, Itàlia.</field>
</element>
<element name="abstract">
<element name="en">
<field name="value">An increasing amount of digital music is being published daily. Music streaming services often ingest all available music, but this poses a challenge: how to recommend new artists for which prior knowledge is scarce? In this work we aim to address this so-called cold-start problem by combining text and audio information with user feedback data using deep network architectures. Our method is divided into three steps. First, artist embeddings are learned from biographies by combining semantics, text features, and aggregated usage data. Second, track embeddings are learned from the audio signal and available feedback data. Finally, artist and track embeddings are combined in a multimodal network. Results suggest that both splitting the recommendation problem between feature levels (i.e., artist metadata and audio track), and merging feature embeddings in a multimodal approach improve the accuracy of the recommendations.</field>
</element>
</element>
<element name="provenance">
<element name="en">
<field name="value">Made available in DSpace on 2017-12-18T09:25:49Z (GMT). No. of bitstreams: 1 oramas_deep.pdf: 605732 bytes, checksum: 92a6a376cdeff1cc44e9f235ce544f71 (MD5) Previous issue date: 2017</field>
</element>
</element>
<element name="sponsorship">
<element name="none">
<field name="value">This work was partially funded by the Spanish Ministry of Economy and Competitiveness under the Maria de Maeztu Units of Excellence Programme (MDM-2015-0502).</field>
</element>
</element>
</element>
<element name="format">
<element name="mimetype">
<element name="ca">
<field name="value">application/pdf</field>
</element>
</element>
</element>
<element name="language">
<element name="iso">
<element name="none">
<field name="value">eng</field>
</element>
</element>
</element>
<element name="publisher">
<element name="ca">
<field name="value">ACM Association for Computer Machinery</field>
</element>
</element>
<element name="relation">
<element name="ispartof">
<element name="none">
<field name="value">DLRS 2017. 2nd Workshop on Deep Learning for Recommender Systems; 2017 Aug 27; Como, Italy. New York: ACM; 2017. p. 32-7.</field>
</element>
</element>
</element>
<element name="rights">
<element name="none">
<field name="value">© 2017 Association for Computing Machinery</field>
</element>
<element name="accessRights">
<element name="none">
<field name="value">info:eu-repo/semantics/openAccess</field>
</element>
</element>
</element>
<element name="title">
<element name="ca">
<field name="value">A deep multimodal approach for cold-start music recommendation</field>
</element>
</element>
<element name="type">
<element name="none">
<field name="value">info:eu-repo/semantics/conferenceObject</field>
</element>
<element name="version">
<element name="none">
<field name="value">info:eu-repo/semantics/acceptedVersion</field>
</element>
</element>
</element>
<element name="subject">
<element name="keyword">
<element name="en">
<field name="value">Recommender systems</field>
<field name="value">Deep learning</field>
<field name="value">Multimodal</field>
<field name="value">Music</field>
<field name="value">Semantics</field>
</element>
</element>
</element>
</element>
<element name="bundles">
<element name="bundle">
<field name="name">THUMBNAIL</field>
<element name="bitstreams">
<element name="bitstream">
<field name="name">oramas_deep.pdf.jpg</field>
<field name="originalName">oramas_deep.pdf.jpg</field>
<field name="description">IM Thumbnail</field>
<field name="format">image/jpeg</field>
<field name="size">19221</field>
<field name="url">http://repositori.upf.edu/bitstream/10230/33519/3/oramas_deep.pdf.jpg</field>
<field name="checksum">41f4813fc6e45e51bd0c8e55fd872fc8</field>
<field name="checksumAlgorithm">MD5</field>
<field name="sid">3</field>
</element>
</element>
</element>
<element name="bundle">
<field name="name">TEXT</field>
<element name="bitstreams">
<element name="bitstream">
<field name="name">oramas_deep.pdf.txt</field>
<field name="originalName">oramas_deep.pdf.txt</field>
<field name="description">Extracted text</field>
<field name="format">text/plain</field>
<field name="size">32866</field>
<field name="url">http://repositori.upf.edu/bitstream/10230/33519/2/oramas_deep.pdf.txt</field>
<field name="checksum">35522f1cee331637b61d25da49f707c4</field>
<field name="checksumAlgorithm">MD5</field>
<field name="sid">2</field>
</element>
</element>
</element>
<element name="bundle">
<field name="name">ORIGINAL</field>
<element name="bitstreams">
<element name="bitstream">
<field name="name">oramas_deep.pdf</field>
<field name="format">application/pdf</field>
<field name="size">605732</field>
<field name="url">http://repositori.upf.edu/bitstream/10230/33519/1/oramas_deep.pdf</field>
<field name="checksum">92a6a376cdeff1cc44e9f235ce544f71</field>
<field name="checksumAlgorithm">MD5</field>
<field name="sid">1</field>
</element>
</element>
</element>
</element>
<element name="others">
<field name="handle">10230/33519</field>
<field name="identifier">oai:repositori.upf.edu:10230/33519</field>
<field name="lastModifyDate">2018-01-24 09:07:37.593</field>
</element>
<element name="repository">
<field name="name">Repositori digital de la UPF</field>
<field name="mail">repositori@upf.edu</field>
</element>
</metadata>