Mailbox search engine using query multi-modal expansion and community-based smoothing

A retrieval method operating on a database of documents including text and names of participants associated with the documents, the retrieval method comprising:

• receiving a retrieval query having at least one of a text query facet comprising one or more keywords and a persons query facet comprising one or more participant names;
• performing query enrichment including at least one of:
• (1) computing an enriched text query as an aggregation of the text query facet, a monomodal expansion of the text query facet based on the one or more keywords, a cross-modal expansion of the text query facet based on the one or more participant names, and a topic expansion of the text query facet based on a topic model associating words and topics, and
• (2) computing an enriched persons query as an aggregation of the persons query facet, a monomodal expansion of the persons query facet based on the one or more participant names, a cross-modal expansion of the persons query facet based on the one or more keywords, and a community expansion of the persons query facet based on a community model associating persons and communities;
• and
• performing ranking including at least one of:
• (1) generating a ranking of documents by sorting similarities between the enriched text query and documents of the database, and
• (2) generating a ranking of persons by sorting the enriched persons query;
• wherein the retrieval method is performed by a computer.

A non-transitory computer-readable storage medium storing instructions executable by a computer to generate a set of topics comprising word distributions and a set of communities comprising participant distributions by performing the non-negative matrix factorizations:

• {
• X
• w
• =
• Θ
• 
• 
• Γ
• t
• 
• Φ
• t
• ,
• Θ
• ≥
• 0
• ,
• Φ
• t
• ≥
• 0
• ,
• Γ
• t
• ≥
• 0
• X
• u
• =
• Θ
• 
• 
• Γ
• c
• 
• Φ
• c
• ,
• Θ
• ≥
• 0
• ,
• Φ
• c
• ≥
• 0
• ,
• Γ
• c
• ≥
• 0
• where Xw is a term-document matrix storing distributions of words in documents of a set of documents, Xu is a participant-document matrix storing participants associated with the documents of the set of documents, Φt is a term-topic distribution matrix storing word distributions for the topics, Φc is a participant-community distribution matrix storing participant distributions for the communities, Θ is a matrix storing the hyper-topic distributions associated with the documents of the set of documents, Γt is a matrix representing the topic distribution per hypertopic, and the Γc is a matrix representing the community distribution per hypertopic.