Amazon.com Inc v. Rensselaer Polytechnic Institute

1. Case Identification

• Case #: IPR2019-01069
• Patent #: 7,177,798
• Filed: May 7, 2019
• Petitioner(s): Amazon.com, Inc.
• Patent Owner(s): Rensselaer Polytechnic Institute
• Challenged Claims: 9-21

2. Patent Overview

• Title: Natural Language Interface
• Brief Description: The ’798 patent relates to a method for processing natural language queries to search a database. The claimed method involves receiving a natural language input, providing a plurality of language-based database objects from that input, identifying permutations of those objects stored in a metadata database, and interpreting a permutation to provide a result.

3. Grounds for Unpatentability

Ground 1: Claims 9-21 are anticipated by or obvious over Livowsky

• Prior Art Relied Upon: Livowsky (Patent 6,598,039).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that Livowsky disclosed every limitation of independent claim 9. Livowsky described a system for searching a database using natural language that receives a user request (limitation 9a) and extracts essential words like "dog" and "sick," along with synonyms like "canine" and "ill," which are the claimed "database objects" (limitation 9b). Petitioner asserted that Livowsky’s system generated ordered and ranked search words (e.g., "Dog Sick," "Canine Ill"), which constituted a "finite number of permutations" (limitation 9c). These objects were stored in Livowsky’s "system database," which Petitioner contended was a "metadata database" because it contained information for interpreting data. This database allegedly included all four required information types: "preference files" that learn from prior searches constituted "case information"; synonyms and user vocabulary were "keywords"; "knowledge trees" and "conceptual graphs" were "information models"; and a "datasoup" of data from the target database constituted "database values" (limitation 9c(ii)). Finally, Livowsky interpreted these permutations by using them to search the database and retrieve results (limitation 9d).
  • Key Aspects: The core of this ground was the assertion that Livowsky’s "system database," with its learning capabilities via "preference files" and its use of knowledge trees, directly read on the ’798 patent’s claimed "metadata database" comprising all four specified information types.

Ground 2: Claims 9-21 are obvious over Shwartz in view of Livowsky

• Prior Art Relied Upon: Shwartz (Patent 5,197,005) and Livowsky (’039 patent).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that Shwartz, a natural language interface for database retrieval, disclosed most limitations of the challenged claims. Shwartz’s "knowledge base" was asserted to be a metadata database. It contained "conceptual information" and an entity-relationship diagram, which Petitioner mapped to "information models" and "graphics." It also stored keywords and database values. Petitioner contended that Shwartz disclosed identifying "candidate navigation paths" (ordered sets of tables and columns) which constituted the claimed "permutations." To the extent Shwartz’s disclosure of "case information" was considered deficient (Shwartz described a manual developer process to refine the knowledge base based on past queries), Petitioner argued it would have been obvious to incorporate Livowsky’s automated learning system.
  • Motivation to Combine: A POSITA would combine Livowsky’s automated learning feature (using "preference files" and "global rules") with Shwartz’s system for several reasons. First, it would automate the manual refinement process described in Shwartz, reducing developer time and ensuring continuous improvement. Second, Livowsky taught that learning from prior searches overcomes the rigidity of conventional systems, a known problem that a POSITA would seek to solve in Shwartz’s system. The combination would predictably result in an improved natural language interface that learns from use.
  • Expectation of Success: A POSITA would have a reasonable expectation of success because using case information to refine a knowledge base was a well-known technique. Both systems were designed for natural language database querying, making the integration of Livowsky’s learning module into Shwartz’s knowledge base a straightforward application of a known technique to improve a similar system.

Ground 3: Claims 9-21 are obvious over Meng in view of Zhang

• Prior Art Relied Upon: Meng (a 1999 technical report) and Zhang (a 1998 PhD thesis).

• Core Argument for this Ground:

  • Prior Art Mapping: Petitioner argued that Meng, which explicitly stated it was an extension of Zhang, disclosed a natural language interface that met all claim limitations. Meng’s system identified keywords from user input that corresponded to nodes, attributes, values, or operators in a "Semantic Graph," which Petitioner asserted was the claimed metadata database. The Semantic Graph itself, being a graphical representation of the database’s semantic model, constituted the "information models" and "graphics." The system used "n-gram vectors" based on "past experiences" to determine the most likely meaning of a keyword, which Petitioner argued was "case information." Zhang, as incorporated by Meng, then used graph search techniques to complete an "incomplete query topic" (derived from the keywords) into a "final connected subgraph," which Petitioner asserted was the claimed "permutation."
  • Motivation to Combine: The motivation was explicit, as Meng stated its interface was an extension of the "High-level Query Formulator" disclosed in Zhang. A POSITA would have combined Meng’s natural language front-end with Zhang’s back-end query formulator precisely as Meng directed.
  • Expectation of Success: A POSITA would have a high expectation of success, as Meng and Zhang described a single, integrated system. The combination involved using known graph search techniques (from Zhang) on a semantic graph representation of a database (from Meng) to predictably generate a valid database query from a natural language input.

• Additional Grounds: Petitioner asserted an additional obviousness challenge against claims 9-21 over Shwartz and Weber, arguing Weber’s teachings on an adaptive system that "learns" words and phrases supplied any missing "case information" element from Shwartz.

4. Key Claim Construction Positions

• "metadata database comprising at least one of a group of information comprising case information, keywords, information models, and database values": Petitioner adopted the district court and prior PTAB construction for this term. For the purposes of the IPR, Petitioner argued this term meant the database must contain all four of the enumerated information types (case information, keywords, information models, and database values). This construction was central to each ground, as Petitioner meticulously argued how each prior art combination disclosed a database containing all four required data types.

5. Relief Requested

• Petitioner requests institution of an inter partes review and cancellation of claims 9-21 of Patent 7,177,798 as unpatentable.