Here's what actually issued. On December 26, 2023, salesforce.com, inc. was granted US11853706B2, "Generative language model for few-shot aspect-based sentiment analysis," inventors Ehsan Hosseini-Asl and Wenhao Liu. The CPC codes are G06F 40/30 (semantic analysis), G06F 40/284, and network classes G06N 3/04, G06N 3/08.
Two modern ideas combine. Aspect-based sentiment analysis doesn't just ask whether a review is positive — it asks how the writer feels about each aspect (the battery, the screen, the service). "Few-shot" means doing this with only a handful of labeled examples rather than thousands. The grant's approach is generative: rather than a classifier head, it uses a generative language model to produce the aspect-sentiment output, which lets it adapt to new aspects with minimal examples.
This is a 2023 grant, so it lands squarely in the generative-AI moment, and the claim reflects it: the invention is using a generative LM for a structured analysis task under low-data conditions. For Salesforce, whose products mine customer feedback at scale, a few-shot method that handles new aspects cheaply is a direct fit for CRM and analytics features.
On scope, the discipline: granted B2, enforceable, but the claims describe a specific generative few-shot method for aspect-based sentiment. They do not cover sentiment analysis broadly, nor generative language models broadly. The boundary is claim 1's allowed language.
The takeaway: US11853706B2 shows the granted record finally catching up to the generative era — a method-specific LLM-based grant for a concrete enterprise task, from a company that productizes exactly this kind of text analysis.