US20130325892A1: Application search query classifier

This article covers the published application as a document: its timeline, what it contains that the granted patent does not, and how to cite it. For the full mechanical breakdown of the invention, see US9405832B2, the granted patent from this family.

Publication number	US20130325892A1
Official title	Application search query classifier
Assignee	Apple Inc.
Inventors	Catherine A. Edwards, Alexander F. Braunstein, Eva H. Mok, Natalia Hernandez-Gardiol
Priority date	May 31, 2012
Filed	August 3, 2012
Published	December 5, 2013
Became	US9405832B2, granted August 2, 2016
Status	Family lapsed for unpaid maintenance fees
International filing	PCT/US2013/038416
Source	patents.google.com/patent/US20130325892A1

What This Publication Covers and Why It Matters for ASO

Feature it describes: App Store search query classification, in its earliest public form.

A published application is not a patent. It is the document a patent office makes public roughly eighteen months after the priority date, before examination has finished and before anyone knows whether the claims will survive. It is worth reading as its own artifact for two reasons that matter to anyone researching app store algorithms.

The first is dating. This publication puts a firm public date on when Apple was working on query intent classification for apps: priority May 31, 2012, which is four years before the granted patent appeared. If you are trying to reconstruct when App Store search stopped being simple keyword matching, the application is the evidence, not the grant.

The second is scope. The specification in a published application usually describes far more than the granted claims end up protecting, and this family is a clear example. The publication describes five categories: navigational, functional, browse, location based and price based. The granted claim in US9405832B2 protects only the navigational and functional split. Everything else Apple described but did not claim is here, in public, and unprotected.

For ASO work the practical takeaways are modest but real:

Query intent classification in the App Store is a 2012 idea, not a recent one. Any strategy still built on the assumption that one ranking function serves all queries has been out of date for over a decade.
The browse category, which covers “best apps” style queries, is described in detail here and dropped from the granted claim. Apple designed for it and chose not to protect it, which tells you something about how central it was considered.
Location based and price based categories appear only as names in a list. That is the weakest kind of patent evidence and should not be cited as if Apple documented how “free” or geographic modifiers are handled.
The international filing shows Apple pursued this across jurisdictions, which is a cost signal. Companies do not file internationally on ideas they consider incidental.

Patent Summary

The application describes a method for responding to app store search queries by first categorizing the query, then choosing a search strategy to match. Three primary categories are defined. Navigational covers searching for a particular app by name, with “Angry Birds” as the example. Functional covers searching for an app that performs a task, with “constellation identification” as the example. Browse covers a user who wishes to explore a variety of apps, with “best apps” as the example. Location based and price based categories are also named.

Categorization draws on the terms in the query, a preliminary search, and empirical data from previous similar searches including the distribution of apps downloaded afterward. The chosen strategy then weights signals differently: navigational searches emphasize exact title matches, while functional searches prioritize metadata similarity and user popularity metrics.

This is the same invention later granted as US9405832B2. The value of reading it in application form is seeing the full scope Apple staked out before examination narrowed it.

Detailed Explanation

What a published application is, and is not

When a company files a US patent application, the application is normally published about eighteen months after its earliest priority date. Publication is automatic and says nothing about quality. It confers no enforceable rights. Examination continues afterward, often for years, and the claims that eventually grant are frequently narrower than the ones filed.

This matters for algorithm research because the two documents answer different questions. The application answers “what was this company working on, and when”. The granted patent answers “what did they successfully claim”. Confusing the two produces the most common error in this field, which is treating a described but unclaimed feature as if it were a protected, and therefore implemented, invention.

The timeline

Date	Event
May 31, 2012	Priority date. The earliest date Apple can claim for this invention
August 3, 2012	Application filed
2013	International application filed as PCT/US2013/038416
December 5, 2013	Published as US20130325892A1
August 2, 2016	Granted as US9405832B2
Later	Family lapsed for unpaid maintenance fees

Put that next to App Store history. In mid 2012 the store held several hundred thousand apps and search was widely criticized as poor. Apple acquired Chomp, the app search startup, in February 2012, three months before this priority date. The timing is circumstantial rather than causal, and the inventor names on this filing are Apple employees rather than Chomp founders, but the sequence is consistent with a company that had just decided app search was a problem worth real engineering.

The full category scheme, as published

The publication lists five categories. Only two of them made it into the granted independent claim.

What survived examination

Category	In the specification	In granted claim 1
Navigational	Described with examples and signals	Yes
Functional	Described with examples and signals	Yes
Browse	Described with examples and signals	No
Location based	Named only	No
Price based	Named only	No

The browse omission is the interesting one. Browse is fully described here, with its own worked example and its own signal set, including the words that point to it: “best”, “top”, “games”, “music”, “video”, “free”, “hot”. It then does not appear in the independent claim that granted. The likeliest explanation is prior art. Distinguishing exploratory browsing from targeted search was already well established in web search literature by 2012, so a claim covering it would have been hard to defend. The narrower navigational versus functional split, where the two are searched on genuinely different bases, is the defensible core.

The lesson generalizes. When a patent family describes more than it claims, the described but unclaimed parts are often the parts that were already obvious. That is a useful filter when reading any of these documents: the claims tell you what was novel, and the specification tells you what the team was thinking about.

What is in the specification but not the claims

Beyond the extra categories, several substantial mechanisms are described in this publication and left out of the granted independent claim. They include the linguistic cue lists, the preliminary search as a classification input, the search term lookup table and its stored fields, the normalized category score comparison, the download distribution analysis with its concentration threshold, and the specific weighting examples such as 0.8 text against 0.2 other for navigational queries.

Some of that appears in dependent claims. The point stands that the richest technical content in this family sits in the specification, which is why the application is worth reading even after the grant exists. Practitioners citing “Apple’s patent” for the download distribution test are, strictly, citing the specification disclosed in this publication.

The international filing

PCT/US2013/038416 was filed in 2013 under the Patent Cooperation Treaty. A PCT filing does not itself grant anything. It preserves the option to pursue national patents in member states, and it costs money at each step.

For research purposes the signal is priority ordering. A company with limited filing budget chooses which inventions get international coverage. Apple chose to extend this one, which puts app store query classification in the tier of work it considered worth defending outside the United States.

What the lapse means

The family ultimately lapsed for unpaid maintenance fees rather than running its full term to the anticipated 2035 expiry. Read that carefully, because it is easy to over interpret.

Letting a patent lapse means the company stopped paying to keep the legal right. It does not mean the technique was switched off. Large portfolios are pruned routinely, and by the time the first maintenance payment came due Apple had filed considerably more specific and more modern work on query understanding. A lapse is a statement about the value of the legal monopoly, not about the codebase.

How to cite this family correctly

To make a claim about what Apple protected, cite US9405832B2 and quote the granted claim.
To make a claim about what Apple described, cite US20130325892A1 and quote the specification.
To make a claim about when, cite the priority date of May 31, 2012, not the grant date of 2016.
Never cite both as if they were two separate inventions. They are one invention at two stages, and counting them twice inflates any tally of Apple’s work in this area.

ASO Patents