Contrastive language and vision learning of general fashion concepts.

The steady rise of online shopping goes hand in hand with the development of increasingly complex ML and NLP models. While most use cases are cast as specialized supervised learning problems, we argue that practitioners would greatly benefit from general and transferable representations of products. In this work, we build on recent developments in contrastive learning to train FashionCLIP, a CLIP-like model adapted for the fashion industry. We demonstrate the effectiveness of the representations learned by FashionCLIP with extensive tests across a variety of tasks, datasets and generalization probes. We argue that adaptations of large pre-trained models such as CLIP offer new perspectives in terms of scalability and sustainability for certain types of players in the industry. Finally, we detail the costs and environmental impact of training, and release the model weights and code as open source contribution to the community.

Draw a Star and Make it Perfect: Incremental Processing of Telicity.

Predicates like "coloring-the-star" denote events that have a temporal duration and a culmination point (telos). When combined with perfective aspect (e.g., "Valeria has colored the star"), a culmination inference arises implying that the action has stopped, and the star is fully colored. While the perfective aspect is known to constrain the conceptualization of the event as telic, many reading studies have demonstrated that readers do not make early commitments as to whether the event is bounded or unbounded. A few visual-world studies tested the processing of telic predicates during online sentence processing, demonstrating an early integration of aspectual and temporal cues. By employing the visual-world paradigm, we tested the incremental processing of the perfective aspect in Italian in two eye-tracking studies in which listeners heard durative predicates in the perfective form in a scenario showing a completed and a non-completed event. Differently from previous studies, we compared telic durative predicates such as "coloring-the-star" to punctual predicates such as "lighting-the-candle." While for punctual predicates, the inferences of telicity (the event has a telos) and of culmination (the telos is reached) are lexically encoded in the perfective verb, for durative predicates, the degree of event completion (visually encoded) needs to be integrated with perfective aspect (linguistically encoded) for the culmination inference derivation. By modulating the interaction of visual and linguistic stimuli across the two experiments, we show that the verb's perfective aspect triggers the culmination inference incrementally during sentence processing, offering novel evidence for the continuous integration of linguistic processing with real-world visual information.

Shopper intent prediction from clickstream e-commerce data with minimal browsing information.

We address the problem of user intent prediction from clickstream data of an e-commerce website via two conceptually different approaches: a hand-crafted feature-based classification and a deep learning-based classification. In both approaches, we deliberately coarse-grain a new clickstream proprietary dataset to produce symbolic trajectories with minimal information. Then, we tackle the problem of trajectory classification of arbitrary length and ultimately, early prediction of limited-length trajectories, both for balanced and unbalanced datasets. Our analysis shows that k-gram statistics with visibility graph motifs produce fast and accurate classifications, highlighting that purchase prediction is reliable even for extremely short observation windows. In the deep learning case, we benchmarked previous state-of-the-art (SOTA) models on the new dataset, and improved classification accuracy over SOTA performances with our proposed LSTM architecture. We conclude with an in-depth error analysis and a careful evaluation of the pros and cons of the two approaches when applied to realistic industry use cases.