| Patent number | US9104670B2 |
| Official title | Customized search or acquisition of digital media assets |
| Assignee | Apple Inc. |
| Inventors | Andrew Wadycki, Jason Douglas |
| Priority and filing date | July 21, 2010 |
| Granted | August 11, 2015 |
| Status | Active. Anticipated expiry April 27, 2032 |
| Family | US20120023095A1 (published application) |
| Source | patents.google.com/patent/US9104670B2 |
What This Patent Covers and Why It Matters for ASO
Feature it describes: using on device usage data to customize store search results, to decide what content is kept locally, and to acquire content before the user asks for it.
This is the earliest document in the collection that describes personalizing store search by individual behavior, and it is one of only three patents here still active. It was filed in July 2010, which puts it eight years before the priority date of Apple’s app recommendation personalization filing and two years before the query classification family.
Scope note. The subject is digital media assets generally. The worked examples are e-books, games, television episodes and podcasts, and much of the document concerns predictive downloading of sequels and next episodes, which has no clean app equivalent. Apps are one asset type inside the described store rather than the focus. Read it as evidence about Apple’s approach to store search personalization, not as an App Store ranking document.
The part that does transfer is stated plainly in the description: search results can be ranked so that the assets more likely to be of interest to that user, based on their usage data, rank higher. The example given is a user whose usage data shows a preference for television over music seeing TV results above music results for the same query.
What follows:
- Personalized store search is not a recent development in Apple’s thinking. A 2010 filing, still in force, describes ranking search results by an individual’s demonstrated consumption behavior.
- The usage data is collected on the device, continuously, by a component the patent calls a usage monitor. This is behavioral data Apple has and no third party does.
- Ranking can happen either on the device or at the server. The patent says the determination of the customized assets can be performed at either. That flexibility matters, because on device ranking is invisible to any external measurement.
- Depth of engagement is tracked, not just whether something was opened. Progress through a book or a game, and whether content was finished or abandoned, are named. The analogue for an app is obvious even though the patent does not draw it.
- The search interface described includes facet filters, shown in Figures 7A through 7D, which is a reminder that the store’s own filtering surfaces shape what users see before ranking gets a chance to.
Patent Summary
“Improved techniques for using usage data are disclosed. The usage data can pertain to a user of an electronic device. The usage data can be monitored and recorded on the electronic device. … In one embodiment, the usage data can be used to enhance searching of digital assets, such as digital media assets, resident at a local or remote repository. In another embodiment, the usage data can be used to determine content to be resident locally on the electronic device. In still another embodiment, the usage data can be used to present (e.g., display) information on the electronic device.”
US9104670B2, abstract, abridged
Three applications of one input, then. A usage monitor on the device records how the user consumes digital media: what they engage with, how far through it they are, how quickly they progress, whether they finish. That record is used to customize search results, to decide what is worth keeping in local storage, and to pre-acquire related content before the user requests it.
The predictive acquisition path is the most developed in the document. Assets carry position markers described as near end marks. When the user crosses one, the system identifies related assets, such as the next episode or a sequel, and downloads them in the background so they are available without network access when the user reaches the end.
The full claim text is not reproduced in the freely available record through common aggregators, so the analysis below rests on the abstract and description rather than on claim language.
Detailed Explanation
What the usage monitor records
Figure 2A places a usage monitor and a predicted asset manager inside the device alongside the controller, storage and network interface. The monitor runs continuously and writes locally. Four kinds of data are named.
| Data type | What it captures |
|---|---|
| Playback position | Where the user pauses or stops within content that has chapters or levels |
| Consumption pattern | How quickly the user progresses through a book, game, show or podcast |
| Asset interactions | Which media types the user engages with most often |
| Completion status | Whether content is finished or abandoned |
Completion status is the entry that deserves attention. A system that records abandonment has a quality signal independent of ratings, reviews or purchase. Apple’s much later app profile filing names an installation retention factor, and this is the same instinct fourteen years earlier, applied to media rather than to apps.
How search gets customized
Figure 6 lays out the customized search flow: present the interface, receive the request, access the usage data, rank the results, present them.
The ranking rule as described is that results more likely to interest the user, based on usage data, rank higher. The worked example is deliberately simple. A user whose usage data skews toward television sees television results above music results for a query that matches both.
What this describes is personalization at the level of asset type and broad interest rather than at the level of individual titles. That is a coarse form of personalization compared with the user account vectors in the 2025 filings, and the difference is a decent measure of how far the field moved in fifteen years.
Figures 7A through 7D show the search interface itself: a search box, facet filters, results display, and filters applied. Faceted search is a structural constraint that often matters more than ranking. A filter removes candidates outright, and no amount of relevance recovers an asset that has been filtered out.
Where the ranking happens
One sentence in the description carries more weight than its length suggests: the determination of the customized digital assets can be performed at the user computing device or at the online media server.
For anyone trying to measure store behavior from the outside, that is a problem. Server side personalization leaves traces that can be compared across accounts and locations. Device side personalization does not, because the reordering happens after the results leave Apple’s infrastructure. If any of the modern equivalent runs on device, then rank tracking tools are measuring a pre personalization result set, and the gap between what they report and what users see is unmeasurable in principle rather than just in practice.
Predictive acquisition
Figures 3, 4A and 4B cover the acquisition path, and Figure 2B shows an asset with its playback position and trigger points marked.
The mechanism: the device watches for the user crossing a near end mark, positioned before the conclusion of the content. On crossing it, the system determines related assets, retrieves them over whatever network connectivity is available, and stores them in a dedicated portion of local storage that refreshes as interests change. The user then receives the content without waiting and without needing a connection, with a purchase option presented in the flow.
The examples are concrete. A second e-book by the same author pre-loads as the first nears its end. A game sequel or expansion downloads before the current game is completed. The next podcast episode is pre-stored in advance of any user request.
There is no honest app equivalent of this, and pretending otherwise would be the kind of overreach this site exists to avoid. Apps are not consumed to completion and do not have sequels in the ordinary case. The transferable observation is narrower: Apple has a patented, active design in which the store acquires content on a prediction rather than a request, and the prediction is triggered by on device behavior.
Local storage as a curated cache
The third use of the usage data is deciding what content should be resident locally. Predicted assets occupy a dedicated portion of storage which refreshes as the user’s interests evolve.
This turns device storage into a managed shelf rather than a passive record of what the user chose to download. It is worth noting as the same underlying idea that later became offloading unused apps and pre-caching recommendations, though this patent does not describe either of those features.
Why the active status matters
Most of the patents on this site have expired or lapsed. This one is active with an anticipated expiry in 2032, alongside US9280595B2, which runs to 2034, and US8364694B2, which runs to 2028.
Maintained patents are not proof of implementation, and Apple maintains plenty it does not use. But paying to keep a patent alive for twenty two years is a choice, and it is a stronger signal than a lapse in the opposite direction. Of the search personalization documents available, this is the one Apple is still paying for.
Related patents on this site
US20240086412A1 is the app specific successor to the personalization idea, with profiles and clusters instead of raw usage data.
Bir yanıt yazın