Re: detailed comments on the use case and reqs 1.0 draft

At 22:33  +0900 17/04/09, Felix Sasaki wrote:
>Hello Dave, all,
>would somebody object if I try to integrate the (non structural) 
>comments? That seems to be pretty straightforward.

no objections from me!

>2009/4/16 David Singer <<>>
>Hi, after a read-through of 
>I have some detailed comments...
>1:  change 'needs' to 'need'..."In addition, video services on the 
>3:  below the diagram, change 'both' to 'either' (unless somehow the 
>media resource is annotated in both formats at once):  "that will 
>return values from either the XMP or IPTC metadata"
>3:  The last paragraph, the words "applications, like:" and what 
>follows would be better phrased as "applications.  For example:" and 
>then follow with an HTML list (bulleted, probably).
>Requirement r4, I would change "custom" to "user-defined" or 
>"non-standard", I think.
>5.2: towards the end, "etc." is missing its "."
>5.3:  this has the feel of having been edited a few times and the 
>language is now a bit odd.  Can I suggest:
>People nowadays are able to enjoy large number of programs from 
>different content providers (broadcasting companies, Internet video 
>website, etc.). To achieve better user experience, reduce the user's 
>experience of being overloaded, and hence retain users, some systems 
>provide recommendations based on the user's history, ratings, or 
>stated preferences. However, different content providers usually 
>have their specific or proprietary metadata models, which is one of 
>the key problems faced by recommendation service providers. A common 
>ontology spanning different metadata sets can allow recommendation 
>systems to return a better, larger, and more relevant selection than 
>when the metadata systems are unrelated.
>Company A is an IPTV add-value service provider. One of their 
>services is to recommend programs that users might like, based on 
>their watching history or explicit rating of programs. In this 
>system, users are able to watch regular TV programs with electronic 
>program guide (EPG) format metadata, videos such as from YouTube, 
>with website-specific metadata, etc. In order to perform uniform and 
>effective recommendation in the absence of a common set of 
>vocabularies, they would need to design own integrated media 
>annotation model.
>5.5:  the set over which "Find all" is not well identified, I assume 
>it's "within a database, such as that of a search engine indexing 
>the internet or other web-accessible content (e.g. a corporate 
>repository, library, etc.)".
>5.7:  I think there is a major use case that needs mentioning: 
>accessibility.  There are requirements that for users who are unable 
>to consume time-based media in general, or some formats in 
>particular, the media data have annotations and links that express a 
>summary, transcript, and so on.
>6:  Requirements.  Very little is said here about the format of the 
>returned result.  Most metadata systems are appallingly vague about 
>the format of the stored data (often merely saying it's a string), 
>even when the key suggests a restricted value set (e.g. "creation 
>date").  This should probably be reflected in 6.13 requirement r13, 
>"allow for undefined/unformatted return values for the same 
>property" as well as the current text.
>More structural comments:
>security: we need to say we understand that user-agent access to 
>metadata might give rise to cross-site scripting and other security 
>issues, but that we expect those issues to be handled in the same 
>way as the same issue for images and other embedded data.
>media ID:  we probably want to say that a major question both users 
>and search engines might like to know is whether two pieces of 
>media, two URLs etc. essentially are referring to the same content. 
> This can really only be done with a uniform media identifier 
>system, and unfortunately there is none (despite ISRC etc.).
>time varying:  some metadata is naturally time-varying (e.g. "what 
>chapter or scene of this movie am I in?") and the cue-ranges design 
>of HTML5 is designed to support this (e.g. flipping slides to go 
>with a video of someone speaking).  While I realize that many major 
>metadata systems express time-invariant metadata ('for the whole 
>file'), it might be prudent to document how we intend to handle 
>this. It could be by using URLs that include fragment identifiers in 
>the query (i.e. a script could find a URL on the page and explicitly 
>add "#t=30s" to the end), or it could be by separate arguments to 
>the API.  Some discussion of this captured in the document would be 
>prudent, I think.
>David Singer
>Multimedia Standards, Apple Inc.

David Singer
Multimedia Standards, Apple Inc.

Received on Friday, 17 April 2009 14:07:31 UTC