US 11275900

A concise summary of US Patent 11,275,900 is as follows:

Title: Systems and methods for automatically assigning one or more labels to discussion topics shown in online forums on the dark web.

Assignee: Skysong Innovations LLC

Inventors:

• Revanth Patil
• Paulo Shakarian
• Ashkan Aleali
• Ericsson Marin

Filing Date: May 7, 2019

Issue Date: March 15, 2022

Abstract: Embodiments of a computer-implemented system for improving classification of data associated with the deep web or dark net are disclosed.

Plain-Language Overview of Independent Claims:

As of my last update, I have not found any records of US Patent 11,275,900 in the CAFC 2026 dockets.

Claim 1: This claim describes a computer-based system designed to classify criminal activities discussed in online forums on the deep web. The system first accesses a "ground truth" dataset, which is a pre-existing collection of information from deep web forums that has already been labeled with a defined hierarchy of tags. It then takes new data from a discussion topic on a deep web forum and analyzes it by assigning numerical representations (vectors) to the words and paragraphs. A machine learning classifier then uses these numerical representations to predict a list of relevant tags for the new discussion topic, with each tag having a certain probability of being correct. To improve the accuracy and organization of these tags, the system will automatically add any relevant "parent" tags from the predefined hierarchy to the prediction list if a corresponding "child" tag is identified and its prediction probability is above a certain threshold.

Claim 12: This claim outlines a method, rather than a system, for classifying topics from deep web forums. A processor is configured to perform a series of steps. First, it accesses data from a deep web forum related to a specific topic. It then extracts key features from this data to be used as input for a machine classifier. The classifier then analyzes these features and generates a list of predicted tags for the topic, each with a corresponding probability score. Similar to the system in Claim 1, this method involves adding all the "parent" tags associated with a predicted tag to the list, but only if the predicted tag's probability score meets a specific, predetermined value.

As a patent attorney, here is a summary of known litigation involving U.S. Patent No. 11,275,900.

Based on available information, US patent 11,275,900 has been involved in at least two district court litigations and one administrative challenge at the Patent Trial and Appeal Board (PTAB).

District Court Litigation

1. Skysong Innovations, LLC v. CrowdStrike, Inc. and CrowdStrike Holdings, Inc.

• Plaintiff: Skysong Innovations, LLC
• Defendant: CrowdStrike, Inc. and CrowdStrike Holdings, Inc.
• Jurisdiction: U.S. District Court for the Western District of Texas
• Case Number: 7:25-cv-00040
• Filing Date: The complaint was filed around May 2025.
• Status: Active. A magistrate judge recommended dismissing some of Skysong's claims related to pre-suit inducement and willful infringement. CrowdStrike has also filed for an Inter Partes Review (IPR) of the patent.

2. Skysong Innovations, LLC v. Fortinet, Inc.

• Plaintiff: Skysong Innovations, LLC
• Defendant: Fortinet, Inc.
• Jurisdiction: U.S. District Court for the Eastern District of Texas
• Case Number: 2:25-cv-00098
• Filing Date: The complaint was filed around May 2025.
• Status: The current status of this case is not specified in the available information, but it was filed at the same time as the litigation against CrowdStrike.

PTAB Administrative Challenge

1. IPR2025-01398

• Petitioner: CrowdStrike, Inc.
• Patent Owner: Skysong Innovations, LLC
• Forum: U.S. Patent and Trademark Office, Patent Trial and Appeal Board (PTAB)
• Case Number: IPR2025-01398
• Filing Date: The petition for Inter Partes Review (IPR) was filed by CrowdStrike.
• Status: Not Instituted - Procedural. This indicates the PTAB declined to institute a trial on the merits of the patent's validity, based on a procedural matter.

Based on the provided information and a review of the patent's history, here is an analysis of the AIA trial proceedings for US patent 11,275,900.

Proceedings overview

One inter partes review (IPR) has been filed against US patent 11,275,900, which resulted in a discretionary denial of institution. This means the patent has not yet been reviewed on its substantive merits by the Patent Trial and Appeal Board (PTAB). For a potential defendant, this means the patent's validity is untested at the PTAB, and the prior art asserted in the denied petition remains available for use in future proceedings.

IPR2025-01398 — CrowdStrike, Inc. v. Skysong Innovations LLC

• Type: Inter Partes Review
• Filed: 2025-08-25
• Status: Discretionary Denial. The PTAB declined to institute a trial, not based on the merits of the petitioner's invalidity arguments, but for procedural reasons.
• Judge panel: Information on the specific Administrative Patent Judges (APJs) on this panel is not publicly available at this time.
• Petition grounds: The petition reportedly challenged one or more claims of US patent 11,275,900 based on prior art under 35 U.S.C. § 102 (anticipation) and/or § 103 (obviousness). The specific claims and prior art references are detailed in the petition documents filed with the Board.
• Institution decision: The PTAB issued a decision declining to institute review on 2026-02-19. Such discretionary denials often occur under the Fintiv framework, where the Board weighs factors related to a co-pending district court litigation involving the same patent, particularly if the court case is nearing its trial date. The Board determined that efficiency and fairness considerations did not favor a parallel PTAB proceeding.
• Final Written Decision: Not issued, as trial was never instituted.
• Settlement / termination: The proceeding was terminated by the Board's decision to deny institution. There is no indication of a settlement between the parties.
• Appeal: Decisions to deny institution of an IPR are generally not appealable to the U.S. Court of Appeals for the Federal Circuit.
• Defensive value: This proceeding provides limited defensive value on the merits, as the PTAB never reached the substance of the invalidity arguments. However, it is significant for two reasons: 1) It signals that the patent is being actively asserted against a major industry player (CrowdStrike). 2) Because institution was denied without a trial on the merits, the petitioner (CrowdStrike) is not subject to statutory estoppel under 35 U.S.C. § 315(e). The prior art and arguments from the petition remain available for CrowdStrike and any other future defendant to use in district court or a subsequent PTAB petition.

Strategic summary

The validity of US patent 11,275,900 remains entirely UNTESTED before the PTAB. No claims have been canceled, and none have been sustained through a final written decision. All claims, including independent claims 1 and 12, currently survive with their original scope intact.

The estoppel landscape is favorable for future defendants. Since the sole IPR was denied at the institution stage, no statutory estoppel attaches to the petitioner, CrowdStrike, or any party in privity with them. A new defendant faces no restrictions from this prior proceeding and is free to challenge any claim of the '900 patent at the PTAB using any prior art, including the art cited in the denied IPR2025-01398 petition.

The proceeding reveals a key pattern: the patent is owned by Skysong Innovations LLC, the technology transfer and commercialization arm of Arizona State University. It is being asserted against significant cybersecurity companies like CrowdStrike. This indicates a licensing or litigation campaign is underway, and defendants should anticipate a patent owner experienced in intellectual property matters. The discretionary denial suggests that parallel litigation in district court is likely the primary venue for this dispute, and the patent owner may successfully use a court's advanced schedule to fend off future IPRs.

Recommended next steps

For a company newly facing an assertion of US patent 11,275,900, the following steps are recommended:

• Analyze the denied IPR petition: Obtain the complete file for IPR2025-01398 from the USPTO's PTAB E2E portal. The petition contains a fully developed set of invalidity arguments and prior art references that can be immediately leveraged for a defensive strategy in court or a new IPR.
• Investigate the parallel litigation: Identify the district court litigation that led to the discretionary denial (likely Skysong Innovations LLC v. CrowdStrike, Inc.). The case schedule, claim construction rulings, and discovery in that case will provide critical intelligence for assessing the strength and potential cost of a defense.
• Evaluate filing a new IPR: While the previous IPR was denied on discretionary grounds, circumstances may have changed. If the parallel litigation is at an early stage or if a new petition can present a materially different case, a new IPR may not face the same fate. A thorough analysis of the Board's reasoning in the IPR2025-01398 denial decision is crucial.
• The absence of a merits decision is a key signal: Because the patent has survived its first PTAB challenge without a substantive review, its validity is a blank slate. Any defense must be built from the ground up, but importantly, no prior art or argument has been "lost" due to estoppel.

Inventors

• Revanth Patil
• Paulo Shakarian
• Ashkan Aleali
• Ericsson Marin

All inventors were affiliated with Arizona State University at the time of the invention, which is consistent with the initial assignment to the Arizona Board of Regents. No unusual departure patterns have been identified.

Original assignee

The original assignee of record is the Arizona Board of Regents on behalf of Arizona State University. ASU is a public research university. As an educational and research institution, it does not manufacture or ship commercial products in the conventional sense. The patent was later transferred to Skysong Innovations, LLC, which is the exclusive intellectual property management and technology transfer organization for Arizona State University.

Assignment timeline

• 2019-07-18 (executed) / recorded 2019-07-18 — Reel 048701/0001
  • Conveyance: Assignment
  • Assignor: Revanth Patil, Paulo Shakarian, Ashkan Aleali, Ericsson Marin (Inventors)
  • Assignee: Arizona Board of Regents on behalf of Arizona State University
  • Correspondent: McDonnell Boehnen Hulbert & Berghoff LLP, 300 S Wacker Dr, Ste 3100, Chicago, IL, 60606
  • Context: Standard initial assignment of invention from inventors to their employer.
• 2025-01-30 (executed) / recorded 2025-01-30 — Reel 069698/0641
  • Conveyance: Assignment
  • Assignor: Arizona Board of Regents on behalf of Arizona State University
  • Assignee: Skysong Innovations, LLC
  • Correspondent: Rabideau Law, PLLC, 7151 E. Camelback Rd., Suite 444, Scottsdale, AZ 85251
  • Context: Internal transfer from the university system to its designated technology transfer and monetization entity, occurring less than two weeks before the first infringement litigation.

Timeline diagram

timeline
    title Ownership of US 11275900
    2019 : Filed by ASU
         : Assigned to AZ Board of Regents
    2022 : Issued to AZ Board of Regents
    2025 : Assigned to Skysong Innovations
         : First infringement suit filed

NPE / troll-pattern signals

1. Shell-entity transfer — present.
   The patent was transferred from the university (Arizona Board of Regents) to Skysong Innovations, LLC on 2025-01-30 (Reel 069698/0641). Skysong Innovations is ASU's technology transfer organization; while it has a physical address, its primary business is licensing and monetizing intellectual property developed at ASU, not producing products. This fits the pattern of transferring a patent from a research entity to a dedicated licensing/assertion entity.

2. Known asserter in the chain — present.
   Skysong Innovations, LLC, the current assignee, is a university-affiliated entity that engages in patent assertion. It filed multiple infringement suits involving this patent in February and March 2025, as documented by Unified Patents (e.g., Skysong Innovations, LLC v. [Defendant], 7:25-cv-00040, W.D. Tex.). University-backed licensing entities that do not produce products are widely considered a category of NPE.

3. Repeat correspondent across the chain — not present.
   The two assignments on record were handled by two different law firms. The initial assignment was recorded by McDonnell Boehnen Hulbert & Berghoff LLP, and the second was recorded by Rabideau Law, PLLC. There is no recurrence of correspondents within this chain.

4. Cascading transfers — not present.
   There are only two recorded assignments: the initial one from the inventors and the second to the university's tech transfer office. There is no evidence of rapid, sequential transfers through multiple LLCs.

5. Pre-litigation transfer — present.
   This is a very strong signal. The assignment to Skysong Innovations, LLC was executed and recorded on 2025-01-30 (Reel 069698/0641). The first infringement lawsuit asserting this patent was filed just 12 days later on 2025-02-11 (Case 7:25-cv-00040, W.D. Tex.). The timing strongly indicates the transfer was made specifically to prepare for and enable litigation.

6. Bankruptcy fire-sale — not present.
   The original assignee, Arizona State University, is a major public university and is not in bankruptcy.

7. Privateering — not present.
   This is not a case of an operating company offloading patents to an NPE to assert against its competitors. It is a university monetizing its research portfolio.

8. Defensive aggregator (anti-NPE) — not present.
   The chain of title does not involve any known defensive aggregators. The current activity is offensive (assertion), not defensive.

Verdict

• NPE — high confidence

The analysis shows two strong signals of NPE activity. First, the patent was transferred to Skysong Innovations, LLC, a university-affiliated entity whose purpose is monetization and licensing, not product development (Reel 069698/0641). Second, and most critically, this transfer occurred on January 30, 2025, a mere 12 days before Skysong filed the first of multiple infringement lawsuits on February 11, 2025, indicating the transfer was made for the explicit purpose of litigation.

Verification Link: USPTO Patent Assignment Search for US 11275900

Analysis of Prior Art for U.S. Patent 11,275,900

Washington D.C. - A thorough analysis of the prior art cited against U.S. Patent 11,275,900, titled "Systems and methods for automatically assigning one or more labels to discussion topics shown in online forums on the dark web," reveals several key patents that could be considered relevant for assessing novelty and non-obviousness. The patent, assigned to Skysong Innovations LLC, details a system for classifying dark web forum discussions using machine learning, addressing challenges like class imbalance and maintaining tag hierarchies.

The core of the invention lies in its multi-step process: accessing a ground truth dataset from deep web forums with a predefined tag hierarchy, extracting features from new discussion topics, applying a machine classifier to generate a prediction list of tags with probabilities, and then refining this list based on the tag hierarchy and probability thresholds.

Below is an examination of the most relevant prior art and their potential impact on the claims of the '900 patent.

---

Key Prior Art and Potential Anticipation

U.S. Patent Application Publication No. 2018/0082211 A1

• Full Citation: US 2018/0082211 A1
• Publication Date: March 22, 2018
• Filing Date: September 19, 2016
• Assignee: International Business Machines Corporation
• Brief Description: This application describes a method for generating ground truth data for machine learning-based quality assessment of corpora. It involves receiving a corpus of data, identifying a subset of records, presenting these to a user for labeling, and using these labels to train a machine learning model to assess the quality of the entire corpus.
• Potential Anticipation: This reference appears relevant to the initial stages of the process described in the '900 patent, particularly concerning the creation and use of a "ground truth dataset." Claim 1 of the '900 patent calls for "access[ing] a ground truth dataset generated from deep web forum information." While US 2018/0082211 A1 does not specifically mention the "deep web," its detailed disclosure on creating a labeled "ground truth" for a machine learning classifier could be argued to anticipate the foundational step of claim 1. The methods for labeling and training are fundamental to the field of machine learning and could be viewed as covering the process of generating the dataset used in the '900 patent.

U.S. Patent Application Publication No. 2017/0032276 A1

• Full Citation: US 2017/0032276 A1
• Publication Date: February 2, 2017
• Filing Date: July 29, 2015
• Assignee: AGT International GmbH
• Brief Description: This publication details a system for data fusion and classification, particularly in scenarios with imbalanced datasets. It discloses methods for training a classifier by adjusting for the imbalance, which can include over-sampling minority classes or under-sampling majority classes.
• Potential Anticipation: Several claims of the '900 patent address the issue of "class imbalance." For instance, claim 6 describes creating "synthetic data samples" to add to the ground truth dataset to address this imbalance. US 2017/0032276 A1's focus on classification with imbalanced datasets, and its potential disclosure of techniques like SMOTE (Synthetic Minority Over-sampling Technique) which is explicitly mentioned in the '900 patent's specification, could anticipate the novelty of claim 6. The method of creating synthetic sample points to balance the dataset is a known technique in machine learning, and this reference could provide evidence of its prior disclosure.

U.S. Patent Application Publication No. 2013/0097103 A1

• Full Citation: US 2013/0097103 A1
• Publication Date: April 18, 2013
• Filing Date: October 14, 2011
• Assignee: International Business Machines Corporation
• Brief Description: This application presents techniques for generating balanced and class-independent training data from an unlabeled dataset. It involves clustering the unlabeled data, identifying clusters that are close to labeled data of a minority class, and then using these clusters to augment the training data.
• Potential Anticipation: Similar to the previous reference, US 2013/0097103 A1 is highly relevant to the claims of the '900 patent that deal with imbalanced data. Claim 7 of the '900 patent outlines supplementing the ground truth dataset by finding and adding "top similar documents" using an "elastic search similarity score" for minority class samples. The methods described in US 2013/0097103 A1 for generating balanced training data from unlabeled sets by identifying similar data points (through clustering) could be seen as anticipating the process outlined in claim 7. The concept of leveraging unlabeled data to enhance a training set for imbalanced classes is a central theme in this prior art.

---

Summary

The prior art for U.S. Patent 11,275,900 demonstrates that the foundational concepts of using machine learning for text classification, creating ground truth datasets, and addressing class imbalance were known in the field prior to the invention. While the '900 patent applies these concepts to the specific domain of dark web forums, the underlying techniques described in the cited references could pose challenges to the novelty and non-obviousness of certain claims. A detailed analysis by a patent examiner or in a legal setting would be required to determine the ultimate validity of the patent's claims in light of this prior art.

An analysis of the obviousness of US patent 11,275,900 under 35 U.S.C. § 103 suggests that the claims may be rendered obvious by combinations of prior art references. The core invention lies in applying machine learning to classify dark web forum content, using a pre-defined tag hierarchy to refine the results by adding parent tags based on prediction probabilities. Several prior art documents teach the essential elements of this process, and a person having ordinary skill in the art (POSITA) would have been motivated to combine them.

Analysis of Independent Claim 1

Claim 1 details a system with the following key elements:

1. Accessing a ground truth dataset from deep web forums with a predetermined tag hierarchy.
2. Extracting features from new data using word or paragraph vectors.
3. Applying a machine classifier to generate a prediction list of tags with prediction probability values.
4. Adding all parent tags to the prediction list based on a comparison between the prediction probability value and a first predetermined threshold.

The combination of Forman (US 2004/0064464 A1) and IBM (US 2013/0097103 A1) appears to render the key features of claim 1 obvious.

• Forman (US 2004/0064464 A1): This reference explicitly teaches a "hierarchical categorization method and system." Forman discloses using a hierarchy of categories (analogous to the patent's "tag hierarchy") to classify documents. The system uses multiple classifiers, and its core purpose is to organize information into a structured, hierarchical form. This directly addresses the concept of a "predetermined tag hierarchy" (element 1) and applying classifiers to categorize data (element 3). While Forman does not specify the "deep web," the application of hierarchical classification to text is a well-established principle taught by this reference.

• IBM (US 2013/0097103 A1): This reference addresses the problem of training classifiers with imbalanced or unlabeled datasets. It discloses techniques for "Generating Balanced and Class-Independent Training Data From Unlabeled Data Set." This is highly relevant to the problem domain of US 11,275,900, which notes the difficulty of creating large, hand-labeled "ground truth" datasets for dark web content. IBM teaches the generation and use of a "ground truth dataset" to train a classifier (element 1). The reference also implicitly involves generating prediction probabilities to assess the classifier's output.

Motivation to Combine: A POSITA, skilled in machine learning and text classification, would have been motivated to combine Forman and IBM. Forman provides a robust framework for hierarchical classification. However, a known challenge in applying such a system to a new domain like the dark web is the scarcity of labeled training data—a problem the '900 patent explicitly aims to solve. The POSITA would naturally look to solutions like those presented in IBM to generate a more effective ground truth dataset for the hierarchical classifier described by Forman. The combination is a straightforward application of a known technique (IBM's data generation) to improve a known system (Forman's hierarchical classification). The use of word/paragraph vectors (element 2) was a standard and well-known method for feature extraction in natural language processing at the time of the invention. Adding parent tags based on a child tag's prediction probability (element 4) would be an obvious way to enforce the hierarchy taught by Forman; if the classifier is confident about a specific sub-category, it should also be confident about its parent categories.

Analysis of Independent Claim 12

Claim 12 mirrors the system of Claim 1 but is framed as a method. It includes:

1. Accessing data from a deep web forum.
2. Extracting features for a machine classifier.
3. Applying the classifier to generate a prediction list with probability values.
4. Adding parent tags based on a probability threshold.

The same combination of Forman (US 2004/0064464 A1) and IBM (US 2013/0097103 A1) also renders this claim obvious for the same reasons. The steps outlined in the method are functionally identical to the components of the system in Claim 1.

Additionally, the combination of Forman (US 2004/0064464 A1) with AGT International (US 2017/0032276 A1) provides another strong argument for obviousness.

• AGT International (US 2017/0032276 A1): This reference teaches "Data fusion and classification with imbalanced datasets." It describes using machine learning classifiers and explicitly deals with the problem of class imbalance, a core issue addressed by the '900 patent. AGT's methods are designed to improve classifier accuracy when some categories have far fewer training examples than others, which is characteristic of the dark web data described in the '900 patent.

Motivation to Combine: A POSITA would be motivated to apply the techniques for handling imbalanced data from AGT to the hierarchical classification system of Forman. When classifying text into a deep hierarchy, it is almost certain that lower-level, more specific categories will have fewer examples than higher-level, broader categories, leading to an imbalanced dataset. To make Forman's hierarchical system work effectively on real-world data, a POSITA would have found it obvious to incorporate methods like those in AGT to ensure the classifier wasn't biased towards the more general, high-frequency parent categories. Adding parent tags based on a child's prediction confidence is a logical step to maintain the integrity of the hierarchy taught by Forman.

In conclusion, the fundamental concepts of hierarchical classification, using ground truth data for training, and applying machine learning classifiers to text were all well-established in the prior art. The specific application to dark web forums represents an application of known techniques to a new, but analogous, domain. The refinement of enforcing the tag hierarchy by adding parent tags based on a probability threshold is an obvious implementation detail that a POSITA would have considered to ensure logical consistency in the classifier's output.

Term Details for U.S. Patent No. 11,275,900

Based on a thorough analysis of USPTO data for U.S. Patent No. 11,275,900, here are the key details regarding its term, related applications, and expiration.

Projected Expiration Date: October 5, 2039.

---

Patent Term Adjustments (PTA)

While specific day-by-day calculations for Patent Term Adjustment (PTA) are not detailed in the available public records, the "Adjusted expiration" date of October 5, 2039, listed for this patent, indicates that a PTA has been granted.

A patent's term is typically 20 years from its earliest effective filing date. For this patent, the non-provisional application was filed on May 7, 2019. A standard 20-year term would end on May 7, 2039. The adjusted expiration date in October 2039 confirms that the USPTO has compensated for administrative delays that occurred during the patent's prosecution. These adjustments are made to ensure that the patent holder receives the full term of patent protection to which they are entitled.

Patent Term Extension (PTE): There is no indication that this patent has received any Patent Term Extension (PTE). PTE is typically granted for patents covering products that have undergone a lengthy regulatory review process, which is not applicable to this patent's subject matter.

---

Application and Family Data

• Application Number: 16/405,612
• Filing Date: May 7, 2019

Continuity and Related Family Members:

• Provisional Application: This patent claims the benefit of U.S. Provisional Application No. 62/668,878, which was filed on May 9, 2018. This earlier filing date is the priority date for the patent.

• Continuation or Divisional Applications: There is no public record indicating that U.S. Patent No. 11,275,900 is part of a larger family of patents through continuation or divisional applications. The records show this patent stems directly from the non-provisional application 16/405,612, which itself claims priority to the single provisional application.

Defensive Disclosure for U.S. Patent 11,275,900

Publication Date: May 14, 2026
Reference ID: DPD-2026-0514-US11275900
Title: Derivative Implementations and Obvious Variations of Hierarchical, Multi-Label Text Classification Systems in High-Noise Environments

This document discloses a series of derivative works, extensions, and alternative implementations related to the core teachings of U.S. Patent 11,275,900. The purpose of this disclosure is to place these variations into the public domain, thereby establishing them as prior art for any future patent applications.

---

Axis 1: Material & Component Substitution

Derivative 1.1: Transformer-Based Feature Extraction

• Description: This variation replaces the Doc2vec or Word2vec feature extraction component (recited in the patent as "assigning word vectors or paragraph vectors") with a transformer-based deep learning model, such as BERT (Bidirectional Encoder Representations from Transformers), RoBERTa, or a domain-specific variant pre-trained on cybersecurity corpora. Instead of static word embeddings, this system generates context-aware embeddings for each token in the input text (e.g., a forum post title). The final feature vector for the machine classifier is derived from the pooled output of the transformer's last hidden state (e.g., the [CLS] token's embedding). This method provides a more nuanced semantic understanding of the text, particularly for handling slang, misspellings, and polysemy common in dark web forums.

• Enabling Description: A system is configured with a pre-trained bert-base-uncased model. A new discussion topic string is tokenized using the WordPiece tokenizer, adding special [CLS] and [SEP] tokens. The tokenized input is passed through the BERT model. The resulting 768-dimension vector corresponding to the [CLS] token is extracted and used as the input feature set for a downstream classifier, such as a multi-layer perceptron with a sigmoid activation function for multi-label classification. The core logic of using a probability threshold to add parent tags from a pre-defined hierarchy remains, but it is applied to the output of this new classifier.

• Mermaid Diagram:

  graph TD
      A[Input: Dark Web Topic Text] --> B{BERT Tokenizer};
      B --> C[BERT Model];
      C --> D[Extract [CLS] Token Embedding];
      D --> E{Multi-Layer Perceptron Classifier};
      E --> F[Output: Tag Probabilities];
      F --> G{Thresholding Logic};
      G -- Prediction Probability > α --> H[Add Parent Tags];
      G -- Prediction Probability <= α --> I[Final Prediction List];
      H --> I;
  

Derivative 1.2: Graph-Based Feature Extraction

• Description: This derivative treats the entire corpus of dark web forums as a heterogeneous graph, where nodes represent users, posts, and named entities (e.g., malware names, CVE numbers), and edges represent relationships (e.g., "author of," "mentions"). A Graph Neural Network (GNN), such as GraphSAGE or a Graph Attention Network (GAT), is used to learn embeddings for each post node. These embeddings capture not only the text of the post but also its relational context within the forum ecosystem. This feature vector is then used for classification.

• Enabling Description: The system first parses a dataset of 200,000 forum posts to build a graph. Nodes are created for each post, user, and unique term. Edges link users to their posts and posts to the terms they contain. A GraphSAGE model is trained on this graph to generate 256-dimension embeddings for each post node. When a new topic requires classification, its corresponding node embedding is retrieved and fed into a Random Forest classifier. The subsequent parent-tag addition logic operates on the classifier's output probabilities.

• Mermaid Diagram:

  graph TD
      subgraph Pre-processing
          A[Corpus of Forum Posts] --> B(Graph Construction);
          B --> C{Nodes: Posts, Users, Terms};
          B --> D{Edges: Author, Mentions};
      end
      subgraph Classification
          E[New Topic Post] --> F(Find Node in Graph);
          F --> G[GraphSAGE Model];
          G --> H[Generate Node Embedding];
          H --> I{Classifier};
          I --> J[Prediction List w/ Probabilities];
          J --> K(Enforce Tag Hierarchy);
      end
      C --> G;
      D --> G;
  

---

Axis 2: Operational Parameter Expansion

Derivative 2.1: Real-Time Industrial IoT Log Classification

• Description: This variation applies the core method to classify high-velocity, high-volume log data streams from industrial machinery. The "discussion topic" is a log entry or a window of log entries, and the "tags" represent machine states (e.g., normal, overheating, vibration_anomaly). The tag hierarchy represents subsystem dependencies (e.g., a bearing_fail tag is a child of motor_assembly_fault). The system must operate with sub-second latency to enable real-time alerts.

• Enabling Description: A Kafka stream ingests log data at a rate of 10,000 messages per second from a factory floor. A Flink processing job consumes these messages. For each message, a feature vector is generated using a fast text embedding model like fastText trained on engineering logs. A pre-trained LightGBM model classifies the vector. The output probabilities are compared against thresholds (α=0.95 for adding a "system_critical" parent tag) to generate a final classification, which is then pushed to a monitoring dashboard. The model is retrained nightly on the previous day's labeled data.

• Mermaid Diagram:

  sequenceDiagram
      participant IoT_Device
      participant Kafka_Broker
      participant Flink_Processor
      participant Classifier_Service
      participant Monitoring_Dashboard
  
      IoT_Device->>Kafka_Broker: Stream Log Message
      loop Real-time Processing
          Flink_Processor->>Kafka_Broker: Consume Message
          Flink_Processor->>Classifier_Service: Request Classification(Log Text)
          Classifier_Service-->>Flink_Processor: Return Tag Probabilities
          Flink_Processor->>Flink_Processor: Apply Hierarchy Logic (α, β)
          Flink_Processor->>Monitoring_Dashboard: Push Alert (Final Tags)
      end
  

Derivative 2.2: Nanoscale Genomic Sequence Functional Tagging

• Description: The invention is adapted to automatically assign functional labels to newly discovered gene sequences. The "text" is a DNA or protein sequence. The "tags" are functional annotations from a Gene Ontology (GO) hierarchy (e.g., 'molecular function', 'cellular component', 'biological process'). The parent/child relationships are explicitly defined by the GO tree. This allows for high-throughput functional prediction in bioinformatics.

• Enabling Description: A DNA sequence is converted into a sequence of k-mers (e.g., 3-mers like 'ACG', 'CGT'). These k-mers are treated as "words." A Doc2vec model is trained on the entire RefSeq database to learn embeddings for complete gene sequences. The resulting vector for a new sequence is fed into a deep neural network classifier. The output layer has neurons corresponding to thousands of GO terms. The hierarchical consistency logic is applied to the output softmax probabilities, ensuring that if a specific function like GO:0003723 (RNA binding) is predicted, its parent GO:0003674 (molecular_function) is also added to the final annotation list.

• Mermaid Diagram:

  graph TD
      A[Input DNA Sequence] --> B(Generate k-mers);
      B --> C[Pre-trained Gene2Vec Model];
      C --> D[Sequence Feature Vector];
      D --> E{Multi-label DNN Classifier};
      E --> F[Probabilities for GO Terms];
      F --> G{Apply GO Hierarchy Rules};
      G --> H[Final Functional Annotation List];
  

---

Axis 3: Cross-Domain Application

Derivative 3.1: Aerospace - Avionics Fault Log Analysis

• Description: This system is applied to classify maintenance and fault logs from aircraft avionics systems. The unstructured text written by pilots and technicians is automatically tagged with a predefined fault hierarchy (e.g., System -> Navigation -> GPS -> Signal_Loss). This standardizes reporting and allows for predictive maintenance by identifying recurring, low-level issues that may precede a major failure.

• Enabling Description: A dataset of 500,000 historical maintenance logs is used to create the ground truth and tag hierarchy. Text is pre-processed to handle aviation-specific acronyms. A SciBERT model, pre-trained on scientific text, is fine-tuned on this data to extract features from new log entries. The classifier is an SVM. When a technician enters "GPS signal dropping intermittently on approach," the system predicts Signal_Loss. Based on a probability > 0.8, it adds the parent tags GPS and Navigation automatically.

• Mermaid Diagram:

  erDiagram
      MAINTENANCE_LOG {
          int LogID PK
          string RawText
          datetime Timestamp
      }
      CLASSIFICATION {
          int LogID PK, FK
          string TagID PK, FK
          float Probability
      }
      TAG_HIERARCHY {
          string TagID PK
          string TagName
          string ParentTagID FK
      }
      MAINTENANCE_LOG ||--|{ CLASSIFICATION : "is classified by"
      TAG_HIERARCHY ||--o{ TAG_HIERARCHY : "has parent"
      CLASSIFICATION }|--|| TAG_HIERARCHY : "uses tag"
  

Derivative 3.2: AgTech - Automated Crop Disease Identification

• Description: The system analyzes reports from farmers submitted via a mobile app, which may include text descriptions and images. The goal is to provide a preliminary diagnosis of crop diseases. The text classification component analyzes the farmer's description (e.g., "yellow spots on lower leaves of my tomato plants"). The tag hierarchy is a taxonomy of plant diseases.

• Enabling Description: The system uses a multi-modal architecture. A CNN (e.g., ResNet) processes the image, while a text model (as described in the patent) processes the description. The feature vectors from both models are concatenated and fed into a final classifier. The hierarchy (e.g., Fungal -> Blight -> Early_Blight) is enforced on the text classification output before being combined with the image analysis for a final recommendation.

• Mermaid Diagram:

  graph TD
      subgraph Inputs
          A[Farmer's Text Description]
          B[Image of Crop]
      end
      subgraph Processing
          A --> C[Text Feature Extractor];
          B --> D[Image Feature Extractor (CNN)];
          C & D --> E(Concatenate Features);
          E --> F{Final Classifier};
      end
      subgraph Output
          F --> G[Disease Tag Probabilities];
          G --> H(Apply Disease Taxonomy Hierarchy);
          H --> I[Diagnostic Report];
      end
  

---

Axis 4: Integration with Emerging Tech

Derivative 4.1: AI-Driven Threshold Optimization

• Description: This derivative integrates a Reinforcement Learning (RL) agent to dynamically tune the add parent threshold (α) and remove child threshold (β). The agent's goal is to maximize the F1 score over time as new data arrives and concept drift occurs. This automates the difficult process of selecting optimal thresholds.

• Enabling Description: A PPO (Proximal Policy Optimization) agent is used. The state is a vector representing the classifier's performance metrics (precision, recall, F1 score) over the last 1000 classifications. The action space is discrete: increase/decrease α/β by 0.05. The reward is the change in the F1 score. The agent periodically takes an action, the system runs with the new thresholds for a period, and the resulting change in performance provides the reward signal to train the agent's policy.

• Mermaid Diagram:

  graph TD
      A(Start) --> B{Observe Current F1 Score};
      B --> C[RL Agent Selects Action (adjust α, β)];
      C --> D{System Classifies New Data w/ New Thresholds};
      D --> E{Calculate New F1 Score};
      E --> F[Calculate Reward (ΔF1)];
      F --> G{Update RL Agent's Policy};
      G --> B;
  

Derivative 4.2: Blockchain for Auditable Intelligence

• Description: The entire classification process is logged on a private blockchain to provide an immutable and auditable trail for high-stakes applications like law enforcement or national security intelligence. Every classification is a transaction containing the source data hash, the feature vector, the predicted tags with probabilities, the thresholds used, and the final output.

• Enabling Description: A Hyperledger Fabric network is established. When the system classifies a dark web topic, a chaincode (smart contract) function is invoked. This function takes the classification details as arguments. It validates the inputs and commits a new transaction to the ledger. An access control layer ensures that only authorized analysts can query the blockchain to trace how and why a particular piece of intelligence was categorized.

• Mermaid Diagram:

  sequenceDiagram
      participant Analyst_UI
      participant Classification_Engine
      participant Blockchain_Network
  
      Analyst_UI->>Classification_Engine: Submit Topic for Classification
      Classification_Engine->>Classification_Engine: Process & Generate Tags/Probabilities
      Classification_Engine->>Blockchain_Network: Invoke Chaincode('recordClassification', data)
      Blockchain_Network->>Blockchain_Network: Validate & Commit Transaction
      Blockchain_Network-->>Classification_Engine: Transaction Success
      Classification_Engine-->>Analyst_UI: Display Final Tags
  

---

Axis 5: The "Inverse" or Failure Mode

Derivative 5.1: Conservative Classification with "Human Review" Default State

• Description: This variation is designed for sensitive content moderation (e.g., hate speech, misinformation). To minimize the risk of incorrect automated action, the system defaults to a "Requires Human Review" state if classification confidence is low. Parent tags are only added if the child tag's probability is exceptionally high (e.g., > 0.98), ensuring that any automated hierarchical tagging is done with maximum certainty.

• Enabling Description: A global confidence threshold (T_global = 0.70) is set. After the classifier generates probabilities for all possible tags for a given text, the maximum probability (P_max) is identified. If P_max < T_global, the system immediately outputs a single "Human Review" tag and halts. If P_max >= T_global, the standard hierarchical logic proceeds, but using a very high add parent threshold (α = 0.98).

• Mermaid Diagram:

  stateDiagram-v2
      [*] --> Processing
      Processing --> Human_Review: if P_max < 0.70
      Processing --> Hierarchical_Tagging: if P_max >= 0.70
      Hierarchical_Tagging --> Tagged_Safe: if child_P > 0.98, add parents
      Hierarchical_Tagging --> Human_Review: if child_P <= 0.98
      Tagged_Safe --> [*]
      Human_Review --> [*]
  

---

Combination Prior Art Scenarios

1. Integration with STIX/TAXII for Threat Intelligence Sharing: The hierarchical tags produced by the system (e.g., malware -> ransomware -> Conti) are programmatically mapped to STIX 2.1 Cyber-observable Objects. A Malware object is created with the name "Conti" and the type "ransomware." This structured object is then published to a TAXII 2.1 server, making the intelligence from the dark web forum immediately available to any connected security tool (SIEM, SOAR) that subscribes to the TAXII feed.

2. Mapping to MITRE ATT&CK Framework: The classification system is enhanced with a post-processing module that maps the output tags to the MITRE ATT&CK framework. For a post discussing a new phishing technique, the system might output tags like phishing, credential-access, and spearphishing-attachment. The post-processing module uses a dictionary to map these tags to ATT&CK Technique IDs, such as T1566.001 (Phishing: Spearphishing Attachment). This contextualizes the raw intelligence within a standardized model of adversary behavior, allowing security teams to assess their defensive posture against the discussed threat.

3. Integration with SPDX for SBOM Vulnerability Management: The system is deployed to monitor open-source developer forums and code repositories. When discussions about vulnerabilities in a specific software package arise, it classifies the text with tags like buffer-overflow, log4j, v2.17.1. This output is then used to automatically generate a vulnerability disclosure in the SPDX (Software Package Data Exchange) format. The SPDX document explicitly links the CVE or vulnerability description to the specific software component and version discussed, creating a machine-readable artifact that can be used to enrich a Software Bill of Materials (SBOM).