PTAB

IPR2019-01068

Amazon.com Inc v. Rensselaer Polytechnic Institute

Key Events

Petition

petition

Title: Method for Natural Language Database Searching
Brief Description: The ’798 patent relates to methods for searching a database using natural language queries. It purports to improve upon prior art by allowing "truly natural" queries without rigid templates, using a metadata-driven approach to interpret user input and generate search results.

Prior Art Relied Upon: Livowsky (Patent 6,598,039).
Core Argument for this Ground:
- Prior Art Mapping: Petitioner argued that Livowsky discloses a method for searching a database that accepts "truly natural" queries (e.g., "My dog is sick"). Petitioner contended that Livowsky's "system database" meets the "metadata database" limitation of claim 1 because it stores information for structuring and interpreting data, including all four required information types. Specifically, Livowsky was alleged to disclose: (1) "case information" by learning from prior user searches via "preference files"; (2) "keywords" such as synonyms and related terms used to refer to database objects; (3) "information models" in the form of a "global knowledge tree" creating webs of concepts; and (4) "database values" via a "datasoup" containing a subset of data from the target database.
- Key Aspects: Petitioner emphasized that Livowsky's system learns from user behavior to resolve ambiguity—for example, by associating the term "pound" with "animal shelter" after a user rephrases a query—thereby directly teaching the use of "case information" as construed.

Prior Art Relied Upon: Shwartz (Patent 5,197,005) and Livowsky (Patent 6,598,039).
Core Argument for this Ground:
- Prior Art Mapping: Petitioner asserted that Shwartz describes a natural language database retrieval system whose "knowledge base" functions as the claimed "metadata database." Shwartz was argued to disclose keywords, information models (conceptual information about application databases), and database values. To the extent Shwartz does not explicitly teach "case information," Petitioner argued it would have been obvious to incorporate this feature from Livowsky.
- Motivation to Combine: A Person of Ordinary Skill in the Art (POSITA) would combine Livowsky's learning capabilities (i.e., its use of "case information" via automatically updated preference files) with Shwartz's system to achieve Shwartz's stated goal of ensuring query interpretations properly reflect user intent. Petitioner argued that automating the refinement of Shwartz's knowledge base using Livowsky's methods for learning from prior searches was a predictable improvement over Shwartz's manual refinement process.
- Expectation of Success: A POSITA would have had a reasonable expectation of success in this combination, as adding adaptive learning features to a natural language interface was a known technique for improving system performance and accuracy.

Prior Art Relied Upon: Meng (a 1999 technical report, "Database Query Formation from Natural Language using Semantic Modeling...").
Core Argument for this Ground:
- Prior Art Mapping: Petitioner argued that Meng describes a natural language interface that uses a "Semantic Graph" as its metadata database. This graph was alleged to contain all four required data types: (1) "information models" (the graph itself represents a "web of concepts"); (2) "keywords" (natural language labels attached to nodes and attributes); (3) "database values" (data from the underlying database linked within the graph); and (4) "case information" (using "n-gram vectors" based on "past experiences" to disambiguate keyword meanings). Meng's system was also argued to perform the claimed steps of identifying and combining database objects to interpret a query and present results.
- Key Aspects: Petitioner highlighted that Meng's use of n-gram vectors, which are statistically derived from past user interactions to resolve ambiguity, was a direct disclosure of using "case information" to refine the query processing method.
Additional Grounds: Petitioner asserted additional obviousness challenges, including that claim 6 is obvious over Livowsky in view of Shwartz (Ground 1b), and that claims 1-8 are obvious over Shwartz in view of Weber (Patent 6,532,444) (Ground 3), with Weber being used to supply the "case information" limitation if Shwartz were found deficient in that regard.

"metadata database comprising at least one of a group of information comprising case information, keywords, information models, and database values": Petitioner adopted the prior construction from district court litigation, which requires the metadata database to include all four enumerated types of information. This construction was central to all grounds, as Petitioner systematically mapped the prior art to each of the four distinct types.
"case information": Petitioner adopted the prior construction of "information about prior instances of use of the natural language processing method." The invalidity arguments relied heavily on demonstrating that the prior art systems "learn" from previous user queries or interactions, thereby storing and using information that satisfies this construction.

Petitioner requests institution of an inter partes review and cancellation of claims 1-8 of the ’798 patent as unpatentable.