Hey Erik,

Sorry to be so late to respond, but I wanted to take the time to read your message and the holidays left me a bit scattered.  

Reviewing what you intend to release, I don't see any clear problems with the dataset.   I think the primary concern is really finding PII in the search string itself.  Generally this is due to people accidentally pasting PII into the query box and then having other queries linked to them through some sort of identifier.  

Is there any way (even very confounded) that you could positively identify any content from a query string?  It seems like the LongestMatchSize_mean_query_x_heading could be a strong indicator of content from the query string.  However we don't have anything in article headers that couldn't be innocently placed in a query string.  E.g. let's say we had a header with "Homer Simpson, 742 Evergreen Terrace, 636-555-1024"[1] and someone then searches for that and gets a perfect match, that wouldn't strongly imply that the person searching was Homer Simpson as they could very likely be searching for a legitimate header (non-suppressed or deleted) in Wikipedia.

Either way, I've worked with Jacob Rogers (+CC) to review dataset publications like this in the past (see [2] for an example).  I think he'll have some more specific concerns and advice.  

1. https://en.wikipedia.org/wiki/The_Simpsons_house#Address_and_phone_number -- just so you know I'm not outing someone who might live at this address and phone number :) 

2. https://figshare.com/articles/Deleted_Wikipedia_articles_spam_vandalism_attack_/4245035

-Aaron

On Thu, Dec 22, 2016 at 6:00 PM, Erik Bernhardson <ebernhardson@wikimedia.org> wrote:

tl/dr: Can feature vectors about relevance of (query, page_id) pairs be released to the public if the final dataset only represents query's with numeric id's?

Over the past 2 months i've been spending free time working on investigating machine learning for ranking. One of the earlier things i tried, to get some semblance of proof it had the ability to improve our search results, was port a set of features for text ranking from an open source kaggle competitor to a datset i could create from our own data. For relevance targets I took queries that had clicks from at least 50 unique sessions over a 60 day period and ran them through a click model (DBN). Perhaps not as useful as human judgements but working with what I have available.

This actually showed it has some promise, and I've been moving further along. An idea was provided to me though about releasing the feature vectors from my initial investigation in an open format that might be useful for others. Each feature vector is for a (query, hit_page_id) pair that was displayed to at least 50 users.

I don't have my original data, but I have all the code and just ran through it with 100 normalized queries to get a count, and there are 4852 features. Lots of them are probably useless, but choosing which ones is probably half the battle. These are ~230MB in pickle format, which stores the floats in binary. This can then be compressed to ~20MB with gzip, so the data size isn't particularly insane. In a released dataset i would probably use 10k normalized queries, meaning about 100x this size Could plausibly release as csv's instead of pickled numpy arrays. That will probably increase the data size further, but since we are only talking ~2GB after compression could go either way. 

The list of feature names is in https://phabricator.wikimedia.org/P4677 A few example feature names and their meaning, which hopefully is enough to understand the rest of the feature names:

DiceDistance_Bigram_max_norm_query_x_outgoing_link_1D.pkl

-  dice distance of bigrams in normalized (stemmed) query string versus outgoing links. outgoing links are an array field, so the dice distanece is calculated per item and this feature has the max value.

DigitCount_query_1D.pkl

- Number of digits in the raw user query

ES_TFIDF_Unigram_Top50_CosineSim_norm_query_category.plain_termvec_x_category.plain_termvec_1D.pkl

- Cosine similarity of the top 50 terms, as reported by elasticsearch termvectors api, of the normalized query vs the category.plain field of matching document. More terms would perhaps have been nice, but doing this all offline in python made that a bit of a time+space tradeoff.

Ident_Log10_score_mean_query_opening_text_termvec_1D.pkl

- log base 10 of the score from the elasticsearch termvectors api on the raw user query applied to the opening_text field analysis chain.

LongestMatchSize_mean_query_x_heading_1D.pkl

- mean longest match, in number of characters of the query vs the list of headings for the page

The main question here i think revolves around is this still PII? The exact queries would be normalized into id's and not released. We could leave the page_id in or out of the dataset. With it left in people using the dataset could plausibly come up with their own query independent features to add. With a large enough feature vector for (query_id, page_id) the query could theoretically be reverse engineered, but from a more practical side I'm not sure that's really a valid concern.

Thoughts? Concerns? Questions?

_______________________________________________
AI mailing list
AI@lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/ai