Click Sequences Research Articles

Personalized fund recommendation with dynamic utility learning

This study introduces a fund recommendation system based on the ϵ-greedy algorithm and an incremental learning framework. This model simulates the interaction process when customers browse the web-pages of fund products. Customers click on their preferred fund products when visiting a fund recommendation web-page. The system collects customer click sequences to continually estimate and update their utility function. The system generates product lists using the ϵ-greedy algorithm, where each product on the list has the probability of 1-ϵ of being selected as an exploitation strategy, and the probability of ϵ is chosen as the exploration strategy. We perform a series of numerical tests to evaluate the estimation performance with different values of ϵ.

Cross-lingual Search Intent Understanding Framework Based on Multi-modal User Behavior

This paper proposes a novel cross-lingual search intent understanding framework leveraging multi-modal user behavior analysis. With the increasing complexity of network traffic and the diversity of user behaviors across languages, traditional approaches often struggle to capture and interpret user search intent in multilingual contexts accurately. Our framework integrates multiple behavioral signals, including query patterns, click sequences, and temporal dynamics, through a sophisticated neural tensor network architecture. The system employs a dual-encoder structure with shared parameters to maintain semantic consistency across languages while incorporating a dynamic behavior sequence learning mechanism to capture temporal dependencies. Experimental evaluation was conducted on a large-scale dataset comprising over 6 million user interactions across four language pairs (EN-ZH, EN-ES, EN-FR, EN-DE) collected over six months. The framework significantly improves over baseline methods, demonstrating an average cross-lingual accuracy of 0.923 and behavior prediction precision of 0.891. Ablation studies reveal the critical role of multi-head attention mechanisms and temporal modeling in maintaining system performance. The framework retains real-time processing capabilities with an average latency of 45ms per request under standard load conditions. Our research advances the field of cross-lingual information retrieval by introducing a practical approach to integrating behavioral signals with linguistic features, providing valuable insights for developing more sophisticated multilingual search systems.

Data-driven smoothing approaches for interest modeling in recommendation systems

In recommendation systems, users often click on some items that are distinct from historically clicked items. This verifies the existence of interest gaps between the historical interests reflected by historical behaviors and current interests. The gaps pose key challenges for interest modeling in recommendation systems, i.e., learn user current interests from historical behaviors. To bridge interest gaps, various solutions have been proposed from a model perspective, and mainly focus on capturing interest diversity and interest evolution process (the changes in user interests over time). Different from existing solutions, our key idea is smoothing the historical interests from a data perspective, so that interest gaps can be smoothed and bridged. To realize this idea, we propose two types of data-driven smoothing approaches based on the principle of collaborative filtering, i.e., click-centric smoothing and user-centric smoothing. The approaches can be easily applied to a number of existing solutions and enhance their performances. To ensure that the smoothed interests completely and accurately cover user current interests (coverage and accuracy), the click-centric smoothing leverages the relevant clicks of a click to smooth the interests on the click, and the user-centric smoothing leverages the relevant clicks of a user click sequence to smooth the interests learned from the click sequence. The “smooth” is to enhance the coverage, and “relevant” is to enhance the accuracy. To search the relevant clicks, we develop an iterative Bayesian inference over a click-flow graph, i.e., a new implementation of collaborative filtering. To further enhance the accuracy, we propose new constrained attention networks to generate high-quality smoothing factors from these relevant clicks. Both offline and online experiments show the data-driven smoothing approaches significantly outperform the baselines. The user-centric smoothing has been deployed on the e-commerce platform of Meituan and achieved 1.51% elevation of Click-through & Conversion Rate and 0.53% elevation of Click-through Rate.

A Job Recommendation Model Based on a Two-Layer Attention Mechanism

In the field of job recruitment, traditional recommendation methods only rely on users’ rating data of positions for information matching. This simple strategy has problems such as low utilization of multi-source heterogeneous data and difficulty in mining relevant information between recruiters and applicants. Therefore, this paper proposes a recurrent neural network model based on a two-layer attention mechanism. The model first improves the entity representation of recruiters and applicants through user behavior, company-related knowledge and other information. The entities and their combinations are then mapped to the vector space using one-hot and TransR methods, and a recurrent neural network with a two-layer attention mechanism is used to obtain their potential interests from the click sequence, and then a recommendation list is generated. The experimental results show that this model achieves better results than previous models.

Deep click interest network for reranking hotels

Nowadays, e-commerce platforms of hotels have become a new trend to help people book hotels online. Interest modeling aims to automatically construct user interests that are critical for e-commerce platforms. Although interest modeling has achieved noticeable successes in product recommendations, it has not been explored in hotel search, especially in the reranking stage. This paper studies the interest modeling task for reranking hotels, and identifies two key challenges—high behavior sparsity and large interest gap. To address the challenges, we propose the deep click interest network (DCIN). High behavior sparsity keeps us from extracting rich semantics to characterize user interest preference. Accordingly, we propose the deep click concept to model the multi-view semantics of a click and then enrich the semantics of both a single click and a click sequence, so that the difficulty caused by the behavior sparsity can be addressed. To address the interest gap challenge, DCIN models user interests with two cascaded units: (i) Mutual-attention Interest Calibration Unit (ICU), which uses the candidate hotel to calibrate the embedding of every click, since the candidate hotel is selected by the ranking stage and can represent user current interests to some extent. So the gap between the calibrated embedding and user current interests can be diminished; (ii) Multi-attention Interest Aggregation Unit (IAU), which estimates the weight of a calibrated embedding from multiple perspectives, i.e., context, user feedback and interest consistency. So the calibrated embeddings, which are relevant to the current query context and important to the user, dominate user current interests, and thus the interest gap can be further diminished. Offline experiments over two large datasets and online A/B testing over the platform of Meituan-Hotel show that DCIN significantly outperforms the baselines and improved baselines. Notably, DCIN has been deployed in Meituan-Hotel, resulting in an increase of 2.40% in Click Through Conversion Rate (CTCVR) and that of 1.09% in Click Through Rate (CTR).

A Survey of Sequential Pattern Based E-Commerce Recommendation Systems

E-commerce recommendation systems usually deal with massive customer sequential databases, such as historical purchase or click stream sequences. Recommendation systems’ accuracy can be improved if complex sequential patterns of user purchase behavior are learned by integrating sequential patterns of customer clicks and/or purchases into the user–item rating matrix input of collaborative filtering. This review focuses on algorithms of existing E-commerce recommendation systems that are sequential pattern-based. It provides a comprehensive and comparative performance analysis of these systems, exposing their methodologies, achievements, limitations, and potential for solving more important problems in this domain. The review shows that integrating sequential pattern mining of historical purchase and/or click sequences into a user–item matrix for collaborative filtering can (i) improve recommendation accuracy, (ii) reduce user–item rating data sparsity, (iii) increase the novelty rate of recommendations, and (iv) improve the scalability of recommendation systems.

Trifunctional Dibromomaleimide Reagents Built Around A Lysine Scaffold Deliver Site-selective Dual-modality Antibody Conjugation.

We describe the synthesis and application of a selection of trifunctional reagents for the dual-modality modification of native, solvent accessible disulfide bonds in trastuzumab. The reagents were developed from the dibromomaleimide (DBM) platform with two orthogonal clickable functional groups built around a lysine core. We also describe the development of an aryl diselenide additive which enables antibody disulfide reduction in 4 minutes and a rapid overall reduction-bridging-double click sequence.

How Do Moth Clicks Decrease the Capture Success of Horseshoe Bats?

Moths are among the most diverse and abundant animals. They have evolved a suite of acoustic defenses, such as producing ultrasound to guard against predation by low-duty-cycle bats. Recently, we found that the ultrasonic clicks of several moth species can also help decrease the capture success of high-duty-cycle bats. We tested whether moth clicks could advertise moth toxicity and/or jam the echolocation calls of high-duty-cycle bats. We examined the acoustic response of a hawkmoth (Cechenena minor) and a tiger moth (Creatonotos transiens) to free-flying horseshoe bats, Rhinolophus osgoodi. C. minor and C. transiens produced anti-bat ultrasonic clicks with an average duty cycle of 22.3% and 4.9%, respectively. The clicks of C. minor overlapped spectrally with the echolocation calls of R. osgoodi. C. minor occasionally emitted clicks immediately after the end of bat feeding buzzes. The clicks of C. transiens did not overlap spectrally with the echolocation calls of R. osgoodi. More than half the click sequences of C. transiens were produced before the emergence of bat feeding buzzes and did not overlap temporally with the feeding buzzes. Moreover, we found that C. minor was palatable and C. transiens was unpalatable to R. osgoodi. These results suggest that C. minor might be capable of jamming R. osgoodi and might also use ultrasonic clicks as warning signals. The clicks of C. transiens probably function to advertise moth toxicity. This study highlights the diversity of moth anti-bat sounds and illustrates the potential mechanisms of moth acoustic defense against high-duty-cycle bats.

Quantification of information processing capacity in living brain as physical reservoir

The information processing capacity (IPC) measure is gaining traction as a means of characterizing reservoir computing. This measure offers a comprehensive assessment of a dynamical system's linear and non-linear memory of past inputs by breaking down the system states into orthogonal polynomial bases of input series. In this study, we demonstrate that IPCs are experimentally measurable in the auditory cortex in response to a random sequence of clicks. In our experiment, each input series had a constant inter-step interval (ISI), and a click was delivered with a 50% probability at each time step. Click-evoked multi-unit activities in the auditory cortex were used as the state variables. We found that the total IPC was dependent on the test ISI and reached a maximum at around 10- and 18-ms ISI. This suggests that the IPC reaches a peak when the stimulus dynamics and intrinsic dynamics in the brain are matched. Moreover, we found that the auditory cortex exhibited non-linear mapping of past inputs up to the 6th degree. This finding indicates that IPCs can predict the performance of a physical reservoir when benchmark tasks are decomposed into orthogonal polynomials. Thus, IPCs can be useful in measuring how the living brain functions as a reservoir. These achievements have opened up future avenues for bridging the gap between theoretical and experimental studies of neural representation. By providing a means of quantifying a dynamical system's memory of past inputs, IPCs offer a powerful tool for understanding the inner workings of the brain.

A Graph Positional Attention Network for Session-Based Recommendation

The main idea of a session-based recommendation system is to model the user’s historical click sequence and then summarize user preferences and predict the items the user will interact with. The session recommendation model based on graph neural networks has attracted much attention in recent years because it can accurately obtain the local relationship between items. However, the traditional session recommendation model based on graph neural Networks lack the use of user’s higher-order features or fail to address the impact of item position information on the current session, which are both critical to the recommendation system. In addition, some models proposed the position information while neglects the click frequency information. We propose a graph network recommendation model called GPAN based on position attention in response to the abovementioned problems. Specifically, we propose a novel high-low order session perceptron that uses the perceptron to model undirected and directed graphs separately to obtain high and low order item representations in a session. For position information, we designed a position layer to calculate independently. Finally, the user’s short-term preference and long-term preference are aggregated to obtain the recommendation sequence. The results through a large number of experiments on three real datasets show that the performance of the proposed GPAN model is the best.

Deep Interest Network Based on Knowledge Graph Embedding

Recommendation systems based on knowledge graphs often obtain user preferences through the user’s click matrix. However, the click matrix represents static data and cannot represent the dynamic preferences of users over time. Therefore, we propose DINK, a knowledge graph-based deep interest exploration network, to extract users’ dynamic interests. DINK can be divided into a knowledge graph embedding layer, an interest exploration layer, and a recommendation layer. The embedding layer expands the receptive field of the user’s click sequence through the knowledge graph, the interest exploration layer combines the GRU and the attention mechanism to explore the user’s dynamic interest, and the recommendation layer completes the prediction task. We demonstrate the effectiveness of DINK by conducting extensive experiments on three public datasets.

Intensity and inter-stimulus-interval effects on human middle- and long-latency auditory evoked potentials in an unpredictable auditory context.

It is not known how Auditory-Evoked Responses (AERs) comprising Middle Latency Responses (MLRs) and Long Latency Responses (LLRs) are modulated by stimulus intensity and inter-stimulus interval (ISI) in an unpredictable auditory context. Further, intensity and ISI effects on MLR and LLR have never been assessed simultaneously in the same humans. To address this important question, thirty participants passively listened to a random sequence of auditory clicks of three possible intensities (65, 75, and 85 dB) at five possible ISI ranges (0.25 to 0.5s, 0.5 to 1s, 1 to 2s, 2 to 4s, 4 to 8s) over four to seven one-hour sessions while EEG was recorded. P0, Na, Pa, Nb, and Pb MLR peaks and N1 and P2 LLR peaks were measured. MLRs P0 (p=.005), Pa (p=.021), and Pb (p=<.001) were modulated by intensity, while only MLR Pb (p=<.001) was modulated by ISI. LLR N1 and P2 were modulated by both intensity and ISI (all p values < .001). Intensity and ISI interacted at Pb, N1, and P2 (all p values < .001), with greater intensity effects at longer ISIs and greater ISI effects at louder intensities. Together, these results provide a comprehensive picture of intensity and ISI effects on AER across the entire thalamocortical auditory pathway, while controlling for stimulus predictability. Moreover, they highlight P0 as the earliest MLR response sensitive to stimulus intensity and Pb (~50 ms) as the earliest cortical response coding for ISIs above 250 ms and showing an interdependence between intensity and ISI effects.

A systematic investigation reveals that Ishihara et al.’s (2008) STEARC effect only emerges when time is directly assessed

The Spatial–TEmporal Association of Response Codes (STEARC) effect (Ishihara et al. in Cortex 44:454–461, 2008) is evidence that time is spatially coded along the horizontal axis. It consists in faster left-hand responses to early onset timing and faster right-hand responses to late onset timing. This effect has only been established using tasks that directly required to assess onset timing, while no studies investigated whether this association occurs automatically in the auditory modality. The current study investigated the occurrence of the STEARC effect by using a procedure similar to Ishihara and colleagues. Experiment 1 was a conceptual replication of the original study, in which participants directly discriminated the onset timing (early vs. late) of a target sound after listening to a sequence of auditory clicks. This experiment successfully replicated the STEARC effect and revealed that the onset timing is mapped categorically. In Experiments 2, 3a and 3b participants were asked to discriminate the timbre of the stimuli instead of directly assessing the onset timing. In these experiments, no STEARC effect was observed. This suggests that the auditory STEARC effect is only elicited when time is explicitly processed, thus questioning the automaticity of this phenomenon.

Monitoring-induced entanglement entropy and sampling complexity

The dynamics of open quantum systems is generally described by a master equation, which describes the loss of information into the environment. By using a simple model of uncoupled emitters, we illustrate how the recovery of this information depends on the monitoring scheme applied to register the decay clicks. The dissipative dynamics, in this case, is described by pure-state stochastic trajectories and we examine different unravelings of the same master equation. More precisely, we demonstrate how registering the sequence of clicks from spontaneously emitted photons through a linear optical interferometer induces entanglement in the trajectory states. Since this model consists of an array of single-photon emitters, we show a direct equivalence with Fock-state boson sampling and link the hardness of sampling the outcomes of the quantum jumps with the scaling of trajectory entanglement.

Real-time user clickstream behavior analysis based on apache storm streaming

This paper presents an approach to analyzing consumers’ e-commerce site usage and browsing motifs through pattern mining and surfing behavior. User-generated clickstream is first stored in a client site browser. We build an ingestion pipeline to capture the high-velocity data stream from a client-side browser through Apache Storm, Kafka, and Cassandra. Given the consumer’s usage pattern, we uncover the user’s browsing intent through n-grams and Collocation methods. An innovative clustering technique is constructed through the Expectation-Maximization algorithm with Gaussian Mixture Model. We discuss a framework for predicting a user’s clicks based on the past click sequences through higher order Markov Chains. We developed our model on top of a big data Lambda Architecture which combines high throughput Hadoop batch setup with low latency real-time framework over a large distributed cluster. Based on this approach, we developed an experimental setup for an optimized Storm topology and enhanced Cassandra database latency to achieve real-time responses. The theoretical claims are corroborated with several evaluations in Microsoft Azure HDInsight Apache Storm deployment and in the Datastax distribution of Cassandra. The paper demonstrates that the proposed techniques help user experience optimization, building recently viewed products list, market-driven analyses, and allocation of website resources.

A multi-task learning approach for improving travel recommendation with keywords generation

Travel recommendation is very critical to helping users quickly find products or services that they are interested in. The key to travel recommender systems is learning user shopping intentions, which are expressed through various supervision signals, such as the clicked products and their titles. Existing travel recommendation methods commonly infer user intentions from click behaviors on travel products. However, remarkable keywords in the product title, such as departure, destination, travel time, hotel, and transportation are paid less attention. To this end, we hypothesize that modeling click sequences and product keywords in title jointly would result in a more holistic representation of a product and towards more accurate recommendations. Thus, we propose a TRKG (short for Travel Recommendation with Keywords Generation) model, which fulfills the travel recommendation and keywords generation tasks simultaneously. To generate explainable outputs, unlike most previous approaches that regard the product title as a hidden feature vector, TRKG regards keywords in the product title as an additional supervision signal. Meanwhile, TRKG integrates the long-term and short-term user preferences in the travel recommendation component and the keywords generation component. To evaluate the proposed model, we constructed datasets from a large tourism e-commerce website in China. Extensive experiments demonstrate that the proposed method yields significant improvements over state-of-the-art methods.

Bionic covert underwater communication focusing on the overlapping of whistles and clicks generated by different cetacean individuals

Bionic camouflage covert underwater acoustic communication technology is playing an increasingly important role in the covert transmission of underwater information. Most cetaceans are group-living animals, and multiple cetaceans often appear in a specific ocean area at the same time. As a result, different cetacean individuals may generate whistles and clicks simultaneously, and the overlapping of whistles and clicks often appears in the received sound signals. Focusing on the overlapping characteristic of cetacean calls, this paper designs a bionic camouflage covert underwater acoustic communication method to improve the covertness and communication rate. The combined encoding through whistles and clicks overlapping of different cetacean individuals makes the constructed communication signals more in line with the characteristics of original cetacean calls, which improves the communication covertness. The pulse-to-pulse correlation and time delay difference encoding based on whistles and clicks are combined to improve the communication rate. Furthermore, the whistle signal processing algorithm based on time–frequency spectrogram image and morphology processing is utilized to separate the whistles and clicks in the received signals. And the whistle and click sequence are decoded, respectively. Finally, the bit error rates of the combined communication sequence, whistle sequence and click sequence are compared by conducting simulations. And the effectiveness, reliability and covertness of the proposed covert communication method are demonstrated through the simulations.

Temporal sensitive heterogeneous graph neural network for news recommendation

News recommendation plays an important role in alleviating information overload and helping users find their interesting news. Most of the existing news recommendation methods make a recommendation based on static data. They ignore the time dynamic characteristics of the interaction between users and news, that is, the order in which users click on news implicitly indicates the user’s interest in news. In this paper, we propose a time sensitive heterogeneous graph neural network for news recommendation. The network consists of two subnetworks. One subnet utilizes convolutional neural network and improved LSTM to learn a user’s stay period on the page and click sequence characteristics as the temporal dimension feature. The other subnet constructs an attention-based heterogeneous graph to model the user-news-topic associations, and apply graph neural network to learn the structural features of the heterogeneous graph as spatial dimensional features. Experiments conducted show that our model outperforms the state-of-the-art models in accuracy and has better interpretability.

Behavioral context of echolocation and prey-handling sounds produced by killer whales (Orcinus orca) during pursuit and capture of Pacific salmon (Oncorhynchus spp.).

Availability of preferred salmonid prey and a sufficiently quiet acoustic environment in which to forage are critical to the survival of resident killer whales (Orcinus orca) in the northeastern Pacific. Although piscivorous killer whales rely on echolocation to locate and track prey, the relationship between echolocation, movement, and prey capture during foraging by wild individuals is poorly understood. We used acoustic biologging tags to relate echolocation behavior to prey pursuit and capture during successful feeding dives by fish‐eating killer whales in coastal British Columbia, Canada. The significantly higher incidence and rate of echolocation prior to fish captures compared to afterward confirms its importance in prey detection and tracking. Extremely rapid click sequences (buzzes) were produced before or concurrent with captures of salmon at depths typically exceeding 50 m, and were likely used by killer whales for close‐range prey targeting, as in other odontocetes. Distinctive crunching and tearing sounds indicative of prey‐handling behavior occurred at relatively shallow depths following fish captures, matching concurrent observations that whales surfaced with fish prior to consumption and often shared prey. Buzzes and prey‐handling sounds are potentially useful acoustic signals for estimating foraging efficiency and determining if resident killer whales are meeting their energetic requirements.

ActionBert: Leveraging User Actions for Semantic Understanding of User Interfaces

As mobile devices are becoming ubiquitous, regularly interacting with a variety of user interfaces (UIs) is a common aspect of daily life for many people. To improve the accessibility of these devices and to enable their usage in a variety of settings, building models that can assist users and accomplish tasks through the UI is vitally important. However, there are several challenges to achieve this. First, UI components of similar appearance can have different functionalities, making understanding their function more important than just analyzing their appearance. Second, domain-specific features like Document Object Model (DOM) in web pages and View Hierarchy (VH) in mobile applications provide important signals about the semantics of UI elements, but these features are not in a natural language format. Third, owing to a large diversity in UIs and absence of standard DOM or VH representations, building a UI understanding model with high coverage requires large amounts of training data. Inspired by the success of pre-training based approaches in NLP for tackling a variety of problems in a data-efficient way, we introduce a new pre-trained UI representation model called ActionBert. Our methodology is designed to leverage visual, linguistic and domain-specific features in user interaction traces to pre-train generic feature representations of UIs and their components. Our key intuition is that user actions, e.g., a sequence of clicks on different UI components, reveals important information about their functionality. We evaluate the proposed model on a wide variety of downstream tasks, ranging from icon classification to UI component retrieval based on its natural language description. Experiments show that the proposed ActionBert model outperforms multi-modal baselines across all downstream tasks by up to 15.5%.