Hyper Search LLC v. Snap Inc

I. Executive Summary and Procedural Information

• Parties & Counsel:
  • Plaintiff: Hyper Search LLC (Texas)
  • Defendant: Snap, Inc. (Delaware/California)
  • Plaintiff’s Counsel: Devlin Law Firm LLC; Toler Law Group, PC
• Case Identification: 1:18-cv-01273, D. Del., 08/20/2018
• Venue Allegations: Venue is alleged to be proper in the District of Delaware because Defendant Snap, Inc. is a Delaware corporation.
• Core Dispute: Plaintiff alleges that Defendant’s Snapchat social media application, specifically its "Discover" screen and underlying content personalization system, infringes three patents related to using neural networks to filter and rank information based on user feedback.
• Technical Context: The technology at issue involves the use of machine learning systems to personalize content feeds, a core functional component of modern social media platforms and digital content delivery services.
• Key Procedural History: The complaint does not mention any prior litigation, inter partes review (IPR) proceedings, or licensing history related to the patents-in-suit.

Case Timeline

Date	Event
1999-02-02	Earliest Priority Date for ’412, ’615, and ’733 Patents
2004-09-14	U.S. Patent No. 6,792,412 Issues
2006-10-10	U.S. Patent No. 7,120,615 Issues
2012-09-04	U.S. Patent No. 8,260,733 Issues
2018-08-20	Complaint Filing Date

II. Technology and Patent(s)-in-Suit Analysis

U.S. Patent No. 6,792,412 - "Neural Network System and Method for Controlling Information Output Based on User Feedback"

• Patent Identification: U.S. Patent No. 6,792,412, issued September 14, 2004.

The Invention Explained

• Problem Addressed: The patent describes a problem arising in the "information age" where users are overwhelmed by vast amounts of available network information. It notes that prior art filtering systems, like those for email, require users to determine filtering attributes in advance, a task that is often difficult and inflexible (’412 Patent, col. 1:36-45).
• The Patented Solution: The invention proposes a system using a neural network interposed between information sources and recipients. This network learns from user feedback about the usefulness of information it delivers and adapts over time. The system operates in "epochs" (e.g., a time period or event), at the end of which it collects feedback, generates a "rating value" for delivered information, and "redetermines the weight values" used by the network to select information for the next epoch (’412 Patent, col. 5:10-24; col. 1:9-15).
• Technical Importance: The technology represents an early approach to applying adaptive, learning-based systems to automate the personalization of content delivered over a computer network, moving beyond static, rule-based filters (’412 Patent, col. 2:10-14).

Key Claims at a Glance

• The complaint asserts at least independent claim 1 (Compl. ¶39).
• Claim 1 requires a system comprising:
  • A plurality of information sources providing information.
  • At least one neural network module that selects objects to receive information based on inputs and weight values.
  • At least one server that provides the selected objects to one or more recipients.
  • Recipients that enable users to generate feedback.
  • The neural network module generates a rating value for objects at the end of an epoch, redetermines weight values using the ratings, and uses the new weights to select objects in a subsequent epoch.
• The complaint does not assert specific dependent claims but reserves the right to modify its infringement theories (Compl. ¶52).

U.S. Patent No. 7,120,615 - "Neural Network System and Method for Controlling Information Output Based on User Feedback"

• Patent Identification: U.S. Patent No. 7,120,615, issued October 10, 2006.

The Invention Explained

• Problem Addressed: As a continuation of the ’412 patent, the ’615 patent addresses the same problem of information overload and the limitations of inflexible, pre-programmed content filters (’615 Patent, col. 1:42-50).
• The Patented Solution: This patent claims a system using a more general "artificial intelligence module" to rank "pieces of information" and place them into "slots" for presentation to a user. The user’s selection of an item from a slot constitutes feedback, which causes the module to "rerank" the information. In response to a "subsequent query," the system presents a modified set of information based on this reranking (’615 Patent, Abstract; col. 20:1-20). This reframes the learning process around the concepts of ranking, slots, and reranking in response to user selection.
• Technical Importance: The invention describes a model for dynamically curating and re-ordering a presented list of content (e.g., links) based on direct user interaction within that list, a foundational concept for personalized feeds (’615 Patent, col. 2:16-20).

Key Claims at a Glance

• The complaint asserts at least independent claim 6 (Compl. ¶71).
• Claim 6 requires a system comprising:
  • A server providing pieces of information to a recipient via a plurality of slots.
  • An artificial intelligence module that ranks the information for placement in the slots.
  • Wherein selection of a slot by the user comprises feedback to the module, causing it to rerank the information.
  • Wherein, in response to a subsequent query, at least one piece of information is either placed in a different slot or not placed at all as a result of the reranking.
• The complaint does not assert specific dependent claims but reserves the right to modify its infringement theories (Compl. ¶84).

U.S. Patent No. 8,260,733 - "Neural Network System and Method for Controlling Information Output Based on User Feedback"

• Patent Identification: U.S. Patent No. 8,260,733, issued September 4, 2012.
• Technology Synopsis: As a continuation of the same patent family, the ’733 patent discloses a system and method for personalizing information output using a neural network. The technology addresses the problem of information overload by learning from user behavior—specifically the selection of links from a list—to modify the weight values in a neural network and subsequently present a more relevant, modified list of links to the user (’733 Patent, Abstract; col. 1:15-21).
• Asserted Claims: The complaint asserts at least independent claim 6 (Compl. ¶103).
• Accused Features: The complaint alleges that Snapchat’s method of presenting stories (links), modifying weight values in a neural network based on user selection or hiding of those stories, learning probabilities of selection, and then conveying a modified list of stories during a subsequent user session infringes the patent (Compl. ¶¶104-108).

III. The Accused Instrumentality

Product Identification

• The accused instrumentalities are the "Snapchat" multimedia messaging application, its "Discover screen," and the "Snapchat neural network system" used for personalizing content (Compl. ¶7).

Functionality and Market Context

• The complaint alleges the accused system provides users with "stories" on the Discover screen. These stories are ranked for display based on a "ranking value" that is determined by implicit user feedback, such as viewing, subscribing to, or sharing the stories (Compl. ¶8, ¶42). The system is also alleged to employ an "artificial intelligence module" or "neural network module" that generates rating values, re-determines rankings based on new interactions, and modifies the content presented in subsequent user sessions (Compl. ¶41, ¶72, ¶77). The complaint further alleges the system allows for explicit user feedback on whether a user wants to see similar stories, which can lead to a particular story being deleted from the user's Discover list (Compl. ¶73).

IV. Analysis of Infringement Allegations

The complaint references external claim chart exhibits (Ex. B, D, F) that were not included in the filing. The infringement analysis is therefore based on the narrative allegations in the body of the complaint.

'412 Patent Infringement Allegations

Plaintiff’s infringement theory for claim 1 of the ’412 Patent is based on mapping the components of the Snapchat platform to the elements of the claimed system. The complaint alleges that Snapchat's user- and friend-generated "stories" constitute the "plurality of information sources" (Compl. ¶40). The "Snapchat neural network module" is alleged to be the claimed "neural network module" that selects these stories for display (Compl. ¶41). Snap's servers are alleged to provide these stories (the "objects") to users' mobile devices (the "recipients") (Compl. ¶45). User interactions (views, likes) are framed as the "feedback," and a "user session" is alleged to be an "epoch." The complaint alleges that at the end of a session, the system generates a "rating value" and "re-determines the ranking values" to control the stories shown in the next session, thereby meeting the learning and redetermination limitations of the claim (Compl. ¶42).

'615 Patent Infringement Allegations

The infringement theory for claim 6 of the ’615 Patent focuses on the ranking and presentation of content. The complaint alleges that Snap's servers provide "pieces of information" (stories) via "slots" on the Discover screen (Compl. ¶72). The "Snapchat artificial intelligence module" is alleged to perform the function of ranking stories for placement in these slots (Compl. ¶72). A user providing feedback—either implicitly through viewing/sharing or explicitly by indicating they do not want to see a similar story—is alleged to be the "act of selection" that constitutes "feedback" (Compl. ¶73, ¶75). This feedback allegedly causes the module to "rerank" the information, resulting in a modified list of stories being presented in a "next user session," which is mapped to the "subsequent query" limitation (Compl. ¶77).

No probative visual evidence provided in complaint.

V. Key Claim Terms for Construction

For the ’412 Patent

• The Term: "epoch"
• Context and Importance: This term is fundamental to the claimed learning cycle. The patent requires the neural network to generate a rating value and redetermine weights "at the end of an epoch" (col. 20:60-67). The viability of the infringement claim will depend on whether Snap's potentially continuous, real-time personalization algorithm can be characterized as operating in the discrete, periodic manner implied by this term. Practitioners may focus on this term because it appears to define a specific batch-processing model that may or may not align with how modern recommendation engines function.
• Intrinsic Evidence for Interpretation:
  • Evidence for a Broader Interpretation: The specification defines an epoch as a "predetermined time, or epoch" or "the conclusion of a predetermined event" (’412 Patent, col. 5:7-8, 5:16-17). This could support the Plaintiff's alleged construction of an epoch as a "user session" (Compl. ¶42), which is an event rather than a fixed time interval for all users.
  • Evidence for a Narrower Interpretation: The specification also describes a "batch prorogation process whereby multiple feedback values are used to recalculate the weight values" (’412 Patent, col. 5:19-21). This language could support a narrower construction requiring a formal, collective processing step where feedback from multiple users or events is aggregated, which may be distinct from the end of a single user's session.

For the ’615 Patent

• The Term: "artificial intelligence module"
• Context and Importance: This term replaced the more specific "neural network module" used in the parent ’412 patent, suggesting a deliberate broadening of scope. Its definition is critical because infringement will depend on whether Snap's proprietary algorithm falls within its scope. Practitioners may focus on this term as it is a key point of differentiation from the parent patent and central to the breadth of the claim.
• Intrinsic Evidence for Interpretation:
  • Evidence for a Broader Interpretation: The patent uses the term "artificial intelligence agent" to describe the invention generally and notes that while a "neural network module may be used," it is an "embodiment" (’615 Patent, col. 3:55-58). The choice to use the broader term in the claim itself, after using a narrower term in the parent, may be evidence of intent to cover a wider range of learning systems beyond just formal neural networks.
  • Evidence for a Narrower Interpretation: The detailed description of the invention, like that of its parent, is almost entirely grounded in the structure and operation of neural networks (e.g., ’615 Patent, col. 6:35-65). A defendant could argue that, when read in light of the specification, the term should be limited to technologies that are structurally or functionally equivalent to the described neural network embodiments, not any algorithm that could be described as "artificial intelligence."

VI. Other Allegations

• Indirect Infringement: The complaint alleges induced infringement for all three patents. The allegations are based on Snap providing "instruction materials, training, and services" and "actively aiding and abetting others to infringe," including partners and customers whose use of the Snapchat system allegedly constitutes direct infringement (Compl. ¶¶48-49, 80-81, 113-114).
• Willful Infringement: The complaint asserts that infringement has been willful "since at least the date Snap received notice by this Complaint" (Compl. ¶¶50, 82, 115). This is a standard allegation of post-filing willfulness, based on knowledge of the patents gained from the lawsuit itself.

VII. Analyst’s Conclusion: Key Questions for the Case

• Technical Operation vs. Claim Language: A central conflict will be whether the actual operation of Snap's commercial-grade, real-time content ranking algorithm can be mapped onto the specific, "epoch"-based batch-processing learning cycle described in the '412 patent. The complaint's success will depend on demonstrating that a "user session" functions as a claimed "epoch" and that the system's continuous updates are equivalent to the claimed "redetermination" of weights.
• Definitional Scope: The case will likely hinge on the construction of key claim terms. The primary question is one of scope: can "epoch" ('412 patent) be interpreted broadly enough to read on a single "user session," and can "artificial intelligence module" ('615 patent) be construed to cover Snap's proprietary ranking algorithm, or will the court adopt narrower definitions tethered more closely to the specific neural network embodiments described in the shared specification?
• The "Subsequent Query" Requirement: For the '615 and '733 patents, a key evidentiary question will be whether the start of a "next user session," as alleged in the complaint, qualifies as the "subsequent query" required by the claims to trigger the presentation of a reranked or modified list of content. This may raise a dispute over whether a passive session initiation constitutes an active "query."