GENERATIVE · 2026-11-16
Generative — Paper Daily, 2026-11-16
Today's generative track fills in what generative recommendation misses when it simply continues a user's history with more SIDs. ReSID finds that long-tail items are mostly pruned by beam search early in decoding, so it keeps tail prefixes alive while decoding. TriGen handles the feed a user enters after clicking an item, letting each request decide how much the click overrides history. TraGR predicts several future steps at once and passes along the candidates it did not choose, weighted by probability. SimRec takes another direction, training the whole search, ask, and recommend loop of a recommendation agent with a simulated user and reinforcement learning, so it learns to check candidates and recover after a rejection.
4.65
LEAD STORY ·
Alibaba /
Nankai University
Generative recommendation encodes each item as a string of Semantic ID tokens and generates them one position at a time with beam search, and cold long-tail items are hard to retrieve this way. The authors take apart the whole chain from encoding through decoding to matching and find that tail items are mostly lost early in decoding: on Amazon Industrial, 62.5% of tail targets have already fallen out of the beam after just two tokens.
4.55
E-commerce has a common kind of recommendation: a user clicks an item in the main feed and lands at once in a feed built around that item. The recommendations must follow the intent of that click without throwing away the user's long-term taste.
4.30
LLM-based recommendation agents can search a catalog, open product details, ask the user, and try another item after a rejection, but when to search, ask, or recommend is mostly a hand-written workflow. Existing training only practices single steps such as rewriting a query or ranking a list, so the outcome of the whole conversation never reaches the policy making those decisions.
4.30
Most recommenders guess only the user's next click, while user trajectory prediction guesses the next several steps at once, such as buying running shoes and then sports socks and a water bottle. Generative recommendation does this by generating each item's SID token by token before moving to the next item, so the encoding within an item and the transitions between items are tangled in one chain.