WebShop: Towards Scalable Real-World Web Interaction with Grounded Language Agents

By Shunyu Yao, Howard Chen, John Yang, Karthik Narasimhan. 2022

1,205 citations recorded by the source corpus.

Existing benchmarks for grounding language in interactive environments either lack real-world linguistic elements, or prove difficult to scale up due to substantial human involvement in the collection of data or feedback signals. To bridge this gap, we develop WebShop -- a simulated e-commerce website environment with $1.18$ million real-world products and $12,087$ crowd-sourced text instructions. Given a text instruction specifying a product requirement, an agent needs to navigate multiple types of webpages and issue diverse actions to find, customize, and purchase an item. WebShop provides several challenges for language grounding including understanding compositional instructions, query (re-)formulation, comprehending and acting on noisy text in webpages, and performing strategic exploration. We collect over $1,600$ human demonstrations for the task, and train and evaluate a diverse range of agents using reinforcement learning, imitation learning, and pre-trained image and language models. Our best model achieves a task success rate of $29\%$, which outperforms rule-based heuristics ($9.6\%$) but is far lower than human expert performance ($59\%$). We also analyze agent and human trajectories and ablate various model components to provide insights for developing future agents with stronger language understanding and decision making abilities. Finally, we show that agents trained on WebShop exhibit non-trivial sim-to-real transfer when evaluated on amazon.com and ebay.com, indicating the potential value of WebShop in developing practical web-based agents that can operate in the wild.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Natural Language Processing, Reinforcement Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Natural Language Processing

In Reinforcement Learning

Citations in the corpus

ReAct: Synergizing Reasoning and Acting in Language Models (2022, background): “We conduct empirical evaluations of ReAct and state-of-the-art baselines on four diverse benchmarks: question answering (HotPotQA, Yang et al., 2018), fact verification (Fever, Thorne et al., 2018), text-based game (ALFWorld, Shridhar et al., 2020b), and webpage navigation (WebShop, Yao et al., 2022”
Reflexion: language agents with verbal reinforcement learning (2023, neutral): “We observe this shortcoming in an experiment on WebShop [29].”
Cognitive Architectures for Language Agents (2023, background): “…2023d), APIs (Schick et al., 2023; Yao et al., 2022b; Parisi et al., 2022; Tang et al., 2023b), and websites (Shi et al., 2017; Nakano et al., 2021; Yao et al., 2022a; Zhou et al., 2023b; Gur et al., 2023; Deng et al., 2023) as well as general code execution (Yang et al., 2023; Le et al., 2022; Ni…”
Understanding HTML with Large Language Models (2022, methodology): “The ﬁrst category uses LLMs to assist web navigation (Nakano et al., 2021; Yao et al., 2022), and typically relies on a custom pre-processing to map the context and structure of a web page to natural language, thus severely restricting what HTML pages the model can parse.”
A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis (2023, methodology): “Evaluation for Real-world Web Navigation Beyond the simulated web environments [59, 81], we have exhibited WebAgent can follow given complex and sometimes ambiguous instructions on the real estate and social media website.”
Mind2Web: Towards a Generalist Agent for the Web (2023, background): “Unlike existing datasets predominantly constructed within simulated environments [27, 36], our objective is to bridge the gap between simulation and reality so that agents trained on our dataset can work on real-world websites out of the box.”
VideoAgent: Long-form Video Understanding with Large Language Model as Agent (2024, neutral): “These models have demonstrated great success across various scenarios, such as online search, card game playing, and database management [25, 26, 61].”
CogAgent: A Visual Language Model for GUI Agents (2023, neutral): “A few prior studies utilized visual features merely as auxiliaries in speciﬁc scenarios. e.g. WebShop [39] which employs visual features primarily for object recognition purposes.”
Tool Learning with Foundation Models (2023, positive): “Through CoT prompting (Self-Ask and ReAct) or fine-tuning (ToolFormer), models can learn to decompose complex questions and utilize the search API to find the answer to the first sub-question.”
ExpeL: LLM Agents Are Experiential Learners (2023, methodology): “…an agent to perform reasoning and question answering using the search tool Wikipedia Docstore API, ALFWorld and WebShop (Shridhar et al. 2021; Yao et al. 2022) that require the agent to perform interactive multi-step decision-making tasks in respectively a household and an online shopping…”
Tool learning with large language models: a survey (2024, neutral): “For example, the employment of search engine tool can enable LLMs to access contemporary information [17, 28, 47–51], while the integration of database tool allows LLMs to access structured databases to retrieve specific information or execute complex queries, thus expanding their knowledge base…”
Auto-GPT for Online Decision Making: Benchmarks and Additional Opinions (2023, methodology): “Webshop [24] is a simulated environment that replicates aweb shopping experience by scraping 1,181,436 products from Amazon.”
ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools (2024, result): “It also demonstrates comparable performance to GPT-4 and Claude 3 Opus in Chinese math and logic reasoning capabilities though it lags behind GPT-4 Turbo.”
Bridging Language and Items for Retrieval and Recommendation: Benchmarking LLMs as Semantic Encoders (2024, neutral): “…LLMs for recommendation can be grouped into three categories: (1) benchmarking their real-world web interaction capabilities as shopping agents (Yao et al., 2022; Wang et al., 2025); (2) benchmarking their shopping-related knowledge and understanding of user intent (Jin et al., 2024); and (3)…”
A survey on large language model based autonomous agents (2023, neutral): “Model Subjective Objective Benchmark Time WebShop [149] 1 2 4 ✓ 07/2022 Social Simulacra [110] 1 3 08/2022 TE [1] 3 08/2022 LIBRO [69] 5 09/2022 ReAct[151] 1 ✓ 10/2022 Out of One, Many[5] 2 2 4 02/2023 DEPS[137] 1 ✓ 02/2023 Jalil et al.”
τ2-Bench: Evaluating Conversational Agents in a Dual-Control Environment (2025, positive): “Following a long line of research into language agents [25, 27, 10, 13, 21], LLM tool use [24, 20, 9], and task-oriented dialog [5, 4, 1, 22, 6, 7, 8], τ -bench [26] is a recently introduced benchmark to measure the reliability of language agents in multi-turn task-oriented conversations such as…”
VisualWebArena: Evaluating Multimodal Agents on Realistic Visual Web Tasks (2024, methodology): “Shopping The Shopping site follows the e-commerce environment from WebArena (Zhou et al., 2024), with product information and content scraped from Amazon and released in WebShop (Yao et al., 2022).”
AgentBench: Evaluating LLMs as Agents (2023, methodology): “We adopt the first 500 entries of 12,087 instructions as test set (following [96]’s official implementation).”