Northern District of California Judge William Alsup has granted summary judgment of noninfringement for Amazon in a case from inventor-controlled MasterObjects, Inc. In a September 15 decision, Judge Alsup held that the plaintiff could not prove literal infringement of an auto-complete search patent under the proper construction of certain key claim limitations, leaving only the issue of infringement under the doctrine of equivalents for possible resolution prior to a scheduled October trial. The ruling follows another, similar setback in one of the first two lawsuits filed in the long-running MasterObjects campaign.
In that early Northern District suit, filed in March 2011 against Alphabet (Google) alongside another short-lived case against Amazon, MasterObjects asserted the three earliest-issuing members of the family at issue (7,752,326; 8,060639; 8,112,529), targeting features in the defendants’ products that suggest completions for a partial text string in an input field. (Only the ‘326 patent was at issue against Amazon.) That initial wave of litigation soon grew to include actions against eBay, Microsoft (dismissed five months after filing), and Yahoo (later acquired by Verizon). However, the plaintiff’s infringement case fell apart in May 2013, when District Judge Phyllis J. Hamilton issued a Markman order that limited the patents’ scope to systems in which just the changes to the string are re-sent back to the server (rather than the new, whole character string). Judge Hamilton then granted summary judgment of non-infringement, and the Federal Circuit subsequently affirmed the district court’s construction. The cases against eBay and Yahoo were stayed during the appeal, and afterward, they were dismissed, Yahoo’s with prejudice and eBay’s without.
In November 2016, MasterObjects sought to salvage its campaign with another Northern District case against eBay, this one asserting for the first time a more recently issued patent (8,539,024) from the same family, again targeting the defendant’s auto-complete features. That litigation was dismissed in light of settlement in December 2018, in the wake of an inter partes review (IPR) in which the Patent Trial and Appeal Board declined to cancel any claims and an Alice motion filed by eBay that had not been decided at the time of dismissal.
MasterObjects then set its sights outside California, suing Amazon in February 2020 in the Southern District of New York and Meta Platforms (then known as Facebook) that March in the Western District of Texas, in both cases asserting the ‘024 patent alongside with three others (9,760,628; 10,311,073; 10,394,866). But that exit was temporary, as the defendants won transfer back to the Northern District of California in October 2020 (Amazon) (3:20-cv-08103) and July 2021 (Meta) (3:21-cv-05428) on convenience grounds.
That litigation this time fell before Judge Alsup, who imposed his unique, distinctive procedure known as a “patent shootout”, or “showdown”, which is designed to more efficiently get to the merits of a patent case by forcing the parties to litigate their best arguments first. In a showdown, the plaintiff must choose a single patent claim for which it thinks that it has the strongest argument and move for summary judgment of infringement and validity as to that sole claim. The defendant, in turn, must pick the weakest claim and move for summary judgment of noninfringement and invalidity. If successful, the plaintiff would possibly be awarded a preliminary injunction; and if unsuccessful, the court might impose sanctions.
However, the Amazon litigation did not move right into the showdown procedure due to a prolonged dispute over the sufficiency of MasterObjects’s infringement contentions. In December 2021, Judge Alsup granted Amazon’s motion to strike the plaintiff’s contentions as to a claim from the ‘628 patent, agreeing with Amazon that the provided claim charts were insufficient due to merely providing screenshots of technical documents without context, but granted leave to amend. MasterObjects would then twice amend its contentions, each time prompting another motion to strike alleging further deficiencies—with Amazon, as it would later recount, asserting that that the plaintiff “still did not set forth a coherent infringement theory” after its second amendment. Judge Alsup denied the latter motion in March 2022 without prejudice for leave to refile at summary judgment.
Those alleged deficiencies identified by Amazon were carried over into the May 2022 report of MasterObjects’s infringement expert, John Peck. In a deposition the following month, per Amazon, Peck admitted that to draft the report, “he copied-and-pasted from MasterObjects’ latest (still deficient) contentions”, the plaintiff’s second amended complaint, and its claim construction briefing, while also improperly incorporating his “understanding from counsel”, failing to review source code, including unverified tests from an “unknown person”, and citing documents that Amazon explained “do not describe the accused system”. In July, Amazon requested leave to file a dispositive summary judgment motion seeking to exclude Peck’s report due to those deficiencies and to resolve “three groups of case-dispositive claim construction disputes”, for which the report failed to address Amazon’s proposed constructions as to those terms. Judge Alsup granted the requested leave, and Amazon filed its motion later that month, addressing the remaining claims from the ‘628, ‘073, and ‘866 patents (as the ‘024 patent was no longer in suit at that point).
Judge Alsup decided that motion on September 15, beginning with an in-depth description of how the accused autocomplete system works. At a high level, he explained that the system processes autocomplete queries using two paired read-only databases (RODBs): a first “Prefix RODB” containing prefixes a user might type to begin a query (“e.g., ‘gold,; ‘golde,’ ‘golden’”) and a second “Keyword RODB” that contains corresponding autocomplete suggestions (e.g., “gold earrings”, “golden girls”, “golden gate bridge”). Matching suggestions are pushed to the user based on the user’s typed prefix. Crucially, though, both RODBs are generated based on Amazon’s “Tommy” search database using “Tommy Query Groups”, which “are aggregated Tommy search-analytics data related to completed product searches” containing “completed search strings” from search requests and related metadata. This distinction was pivotal for the “cache and content source” claim limitations at issue, per Judge Alsup:
This bears emphasizing because the Tommy dataset does not store autocomplete queries and at most stores autocomplete results (and arguably not even that). Rather, the Tommy dataset stores completed product search queries, i.e., not a query where the results are suggestions for a complete query (e.g., “Rol” = “Rolex”), but a query where the corresponding results are listings for actual products Amazon has for sale (e.g., the range of Rolex watches a customer could buy). (Emphasis in original.)
The reason this is so important, he explained, is that the nature of the claimed “cache” limitation is what determines infringement. As summarized by Judge Alsup, Amazon argued that the asserted claims “all require a cache that stores prior autocomplete queries”, whereas “MasterObjects opposes that the claimed system only requires prior autocomplete results and that Amazon’s construction reads limitations from the embodiment into the claims” (emphasis in original). Judge Alsup agreed with Amazon: “Because this order construes the claimed system’s “cache” to store autocomplete queries (as well as results), Amazon’s system does not infringe” (emphasis in original).
The court arrived at this construction based on its interpretation of three key limitations from the independent claims. First, claim 13 specifies that the system both “caches query results” and also has the ability to “look up the query in said cache”, indicating that “the system has also stored the prior query as well”—while the context in which “cache” is used as a verb makes it clear “that data stored in the cache had previously been processed by the data source”. Second, claim 1[e] of the ‘073 patent describes the cached data as “based on content queries received from multiple users”, further supporting the notion that these queries were “already processed by a data source”. Third, and similarly, a limitation from that same claim describing a “cache of queries of search results” that have been “previously retrieved from one or more content sources” clarifies that that those autocomplete queries had been “previously processed by a content/data source”.
Judge Alsup then criticized MasterObjects for changing its claim construction arguments over time as its prior infringement arguments fell apart:
Shifting sands have blown hard in this case. MasterObjects began this litigation expressly in agreement that the claims required a cache of both autocomplete queries and results. As this litigation progressed, however, it has become clear that the accused system does not store queries, so MasterObjects flip-flopped in its opposition to summary judgment, reneging on its prior assertions that the cache terms did indeed require storage of autocomplete queries.
Yet the court noted that its order was consistent with MasterObjects’s original position, noting that this was further supported by the plaintiff’s own expert from an IPR against the ‘073 patent, which it has since “decrie[d]”. Judge Alsup further found that his construction is supported by the specification, including statements that he characterized as providing “insight into how to read the cache limitations” and an embodiment (“QuestObjects”) that, together, “accord[] with the other intrinsic evidence regarding the cache limitations”—notwithstanding MasterObjects’s attempt “to walk back its prior statements with the excuse that its ‘prior vernacular has not always been perfect’”.
In short, Judge Alsup construed “a ‘cache’ as a particular type of memory store that saves a copy of both the autocomplete query and results from the data source”, likewise construing “the verb form of ‘cache’ to mean to store a copy of both the autocomplete query and results from the data source”.
Even MasterObjects’s “best argument”, emphasized Judge Alsup, requires that the prefixes stored in the cache actually correspond to user search queries (e.g., “Rol”) that correspond to a result (e.g., “Rolex”). The problem with this argument, he explained, is that the Prefix RODB is “not a history of prior queries”, but rather “an artificially generated set of all possible prefixes that would lead to ‘Rolex’” (emphasis added). “‘Rol’ in the accused system is not a copy of the autocomplete query but an independently and artificially generated construct. This is sufficient for a finding of noninfringement”. Moreover, Judge Alsup found that “Amazon’s system also does not infringe because the Keyword RODB is built from prior product search queries, not prior autocomplete results like the claimed system”.
The court’s determination that the claimed system “has an underlying data source that can itself be queried” additionally supports a ruling of noninfringement, Judge Alsup held—clarifying that since caching is a “discretionary optimization” designed to reduce search queries from a data source, “the claimed system must be able to query the underlying data source to generate autocomplete query results”. MasterObjects’s arguments to the contrary that turn on the concept of a “cache miss”, the court underscored, depend on the orthogonal “false premise” that the issue is whether the claimed system must “check the data source every time”. But the real requirement that the data source can be queried at all cannot be avoided, per the court: “Since the claimed system’s cache saves a copy of the autocomplete query and the result generated by the data source, the data source itself can inherently be queried”.
As a result, “[b]ecause Amazon’s system does not practice the claimed cache-and-data-source structure recited by the claims-in-suit, there can be no infringement. MasterObjects has accused the RODBs to be the cache in Amazon’s system, which in turn requires the Tommy Query Groups to constitute the system’s data source”. Moreover, the court’s construction of the cache limitations establish that the plaintiff “cannot demonstrate infringement at trial”, particularly since Peck’s opinion provided no “relevant expert testimony”. The order did not address whether Peck’s opinion should be excluded altogether “[b]ecause of this order’s other conclusions”.
All that remained, the court concluded, was the issue of the doctrine of equivalents, which was not addressed in Amazon’s motion. Judge Alsup indicated that this issue may be resolved sooner rather than later, ordering the parties to chart the “proper course” on this issue within five days in statements no longer than five pages—at which point he “may possibly vacate the trial date” currently scheduled for October 11, 2022.
For more on the prior history of this campaign, see “New Suit Against Amazon—and an Answer by Facebook—Filed in Predictive Search Campaign” (May 2020).
