[
https://issues.apache.org/jira/browse/HIVEMALL-74?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15965234#comment-15965234
]
Takuya Kitazawa edited comment on HIVEMALL-74 at 4/12/17 2:22 AM:
------------------------------------------------------------------
There are several incremental variants of pLSA.
One is called *fold-in* which re-runs EM steps for new documents *by fixing
P(w|z)*. This strategy is described in the original papers, but we have no way
to update *P(w|z)* for newly observed documents. So, it does not fit to our
purpose.
Incremental methods which update P(w|z) are described in: "Incremental
Probabilistic Latent Semantic Analysis for Automatic Question Recommendation"
https://pdfs.semanticscholar.org/b66e/c7faf2e4888503f7ad1537d284f350fb3e58.pdf
(Table 1 in the paper compares different algorithms); we will implement one of
the three incremental variants, *IPLSA*, *MAP-PLSA* and *Our method*, in
Hivemall.
was (Author: takuti):
Online PLSA
http://poseidon.csd.auth.gr/papers/PUBLISHED/JOURNAL/pdf/2014/D48.pdf
> Implement pLSA algorithm
> ------------------------
>
> Key: HIVEMALL-74
> URL: https://issues.apache.org/jira/browse/HIVEMALL-74
> Project: Hivemall
> Issue Type: Wish
> Reporter: Makoto Yui
> Priority: Minor
> Labels: Algorithm
>
> Support pLSA [1].
> [1] Thomas Hofmann, "Collaborative Filtering via Gauusian Probabilistic
> Latent Semantic Analysis", Proc. SIGIR, 2003.
> http://dl.acm.org/citation.cfm?id=860483
> http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.7.2476&rep=rep1&type=pdf
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
