So we index the OCR text in two Solr fields. One that is just the OCR text and another that is a accumulation of all of the metadata and OCR for a resource. When a user does a "keyword" search we search against that Solr field with all of the accumulated information. We never use the other Solr field of just the OCR text; expect for troubleshooting. On a hit when a user is directed to a resource if the BookReader is available we only offer a "search inside" option[1]. If the user submits a search using that field then we only search the OCR XML; but we do that outside of Solr. We have a service[2] that returns the payload results for the BookReader to render.
Now back when we started implementing the BookReader when rendering the resource view I queried that same BookReader search service using the keyword supplied by the user that initially got them to the resource. I used it as a kind of look ahead/hint mechanism. I would then sends queues to the interface to let the user know X number of hits were found in the document. I also was able to construct links to the pages in the BookReader where the hits were found. Functionally, it all worked well and was pretty trivial to implement. I tracked usage and asked for feedback. A very high percentage of users just clicked the link to open the BookReader and never used the hinting mechanism. So I disabled it; finding it not worth the extra pull on our system and clients. I never dug deeper to find out what the greater issue was with the low level of use.
Best,
Chad
[1] https://rucore.libraries.rutgers.edu/rutgers-lib/41256/
[2] https://github.com/RutgersUniversityLibraries/OCR-search-for-IA-reader
***************************************************************************
Chad Mills Rutgers University Libraries
Digital Library Architect Scholarly Communication Center
Ph: 848.932.5924 Room 409D, Alexander Library
Fax: 848.932.1386 169 College Avenue, New Brunswick, NJ 08901
Cell: 732.309.8538 https://rucore.libraries.rutgers.edu/
***************************************************************************
----- Original Message -----
From: "Shaun D. Ellis" <[log in to unmask]>
To: [log in to unmask]
Sent: Monday, February 29, 2016 7:13:39 PM
Subject: Re: [CODE4LIB] OregonDigital's BookReader integration (was: Listserv communication)
Josh,
Congrats on the new gig, and thank you for this explanation of OregonDigital’s BookReader integration. I’m sorry I wasn’t more specific about this, but I think the original question had less to do with the BookReader integration, and more to do with a non-frameworky explanation of configuring Solr to return direct links to pages where the keywords appear in a “compound” object, such as a book.
As the original poster (Laura Buchholz) mentioned, it seems like OregonDigital does not provide direct links until after the BookReader is loaded. It’s only then that pins are placed on the “slider nav” to indicate where the keyword appears. So, to answer the original question, it seems like all the full-text may be dumped into a single Solr field that returns the object in the initial search result, and then upon loading the BookReader makes a subsequent query (limited to that one object) retrieve the “data payload” in your example to then locate the exact pages where the terms appear? Is that what’s going on there?
I suppose if you wanted to return all the page numbers in the original search query, you may have to send each page individually to Solr to be indexed, and if you have a viewer with conventions for "deep linking" (like the BookReader has) you could generate the link for each page and index it to provide this functionality.
I was curious as folks were posting all the inspiring digital collections sites earlier today, so I looked for this pattern but didn’t see it. Most of the apps use the same pattern as OregonDigital (although my testing was not particularly thorough, so let me know if I’m wrong, folks!). On the otherhand, you do see the "direct-to-page link" interface with both Amazon and Google Books search, which takes you directly to the page from the initial search results.
So, I’m not sure if this was a conscious design decision on the part of library digital collections creators, if the pattern is followed because it’s considered a “best practice” or a “convention” in our field, or if it was just simpler to implement.
Thanks again for the follow up,
Shaun
> On Feb 26, 2016, at 2:51 PM, Gum, Josh <[log in to unmask]> wrote:
>
> I’m very new (<1 month) to Oregon State University, library technology, and Code4Lib. So please bear with me. Also, I’m going to put a disclaimer out that I may be missing some of the picture here.. I’m willing to lend a hand digging into more details if needed, so please feel free to ask.
>
> Also.. I’m going to split this part of the discussion into a separate thread, so we can address the question regarding the OregonDigital BookReader integration. I’ve done some digging this morning, and spoke to a colleague who took part in some of the text extraction for PDF assets in OregonDigital.. I’m hopeful that these details are enough to help connect the dots regarding our integration.
>
> ————————————
> When ingesting a PDF asset [1], we have a shell based processor [2] which executes “pdftotext” [3] to extract and store the text from a pdf with bounding boxes around each word in the file.
>
> The command executed on the server:
> pdftotext -enc UTF-8 '#{file_path}' '#{output_file}' -bbox
>
> The web UI for viewing a PDF and highlighting results is tied to BookReader [4], which has a great amount of functionality and is well documented online! [5]
>
> The BookReader is making calls to a “full_text” action on the document_controller to find the location of the search terms. [6] This JSONP call to our web server uses OregonDigital::OCR::BookreaderSearchGenerator [7] to supply the properly formatted page and bounding box results to BookReader to use in updating its UI with the appropriate highlights and place marker icons. If you use something like the Chrome DevTools while searching for a term on the BookReader UI, you can see the data payload that is returned from the server. For instance, here’s a snippet of one search I did:
>
>
> (apologies if the tabs don’t remain in the email)
> matches: [
> {
> par: [
> {
> page: 2,
> boxes: [
> {r: 128.62286274509802, l: 101.30935784313726, b: 27.52538962121212, t: 19.953774090909093, page: 2}
> {r: 59.883534313725484, l: 29.41176470588235, b: 242.4078138636364, t: 234.83619833333336, page: 2}
> {r: 106.32754411764705, l: 80.37296078431372, b: 546.3512438560606, t: 538.7796283257576, page: 2}
> text: "McKenzie Highway {{{Historic}}} District…
> }
> ]
> }
> ]
>
>
> [1] https://github.com/OregonDigital/oregondigital/blob/master/app/models/document.rb
>
> [2] https://github.com/OregonDigital/oregondigital/blob/d82d944d55dd087d2670b3f065725ef0e5ddc4ce/lib/hydra/derivatives/pdf_text_processor.rb
> [3] http://www.manpagez.com/man/1/pdftotext/
> [4] http://github.com/openlibrary/bookreader/
> [5] https://openlibrary.org/dev/docs/bookreader
> [6] https://github.com/OregonDigital/oregondigital/blob/master/app/controllers/document_controller.rb
> [7] https://github.com/OregonDigital/oregondigital/blob/master/lib/oregon_digital/ocr/bookreader_search_generator.rb
> ———————————
>
> Josh Gum
> Oregon State University Libraries and Press
>
>
>
>
>
>
> On 2/26/16, 7:07 AM, "Code for Libraries on behalf of Shaun D. Ellis" <[log in to unmask] on behalf of [log in to unmask]> wrote:
>
>> … //SNIPPED
>> I have to admit that I was disappointed that the recent question about full-text searching basics (behind OregonDigital’s in-page highlighting of keywords in the IA Bookreader) went basically unanswered. This was a well-articulated legitimate question, and at least a few people on this list should be able to answer it. It’s actually on my list to try to do it so that I can report back, but maybe someone could save me the trouble and quench our curiosity?
>>
>> Cheers,
>> Shaun
