Click-through Research Articles

Learning to Retrieve User Behaviors for Click-through Rate Estimation

Click-through rate (CTR) estimation plays a crucial role in modern online personalization services. It is essential to capture users’ drifting interests by modeling sequential user behaviors to build an accurate CTR estimation model. However, as the users accumulate a large amount of behavioral data on the online platforms, the current CTR models have to truncate user behavior sequences and utilize the most recent behaviors, which leads to a problem that sequential patterns such as periodicity or long-term dependency are not contained in the recent behaviors but in far back history. However, it is non-trivial to model the entire user sequence by directly using it for two reasons. Firstly, the very long input sequences will make online inference time and system load infeasible. Secondly, the very long sequences contain much noise, thus making it difficult for CTR models to capture useful patterns effectively. To tackle this issue, we consider it from the input data perspective instead of designing more sophisticated yet complex models. As the entire user behavior sequence contains much noise, it is unnecessary to input the entire sequence. Instead, we could just retrieve only a small part of it as the input to the CTR model. In this article, we propose the U ser B ehavior R etrieval (UBR) framework which aims at learning to retrieve the most informative user behaviors according to each CTR estimation request. Retrieving only a small set of behaviors could alleviate the two problems of utilizing very long sequences (i.e., inference efficiency and noisy input). The distinguishing property of UBR is that it supports arbitrary and learnable retrieval functions instead of utilizing a fixed pre-defined function, which is different from the current retrieval-based methods. Offline evaluations on three large-scale real-world datasets demonstrate the superiority and efficacy of the UBR framework. We further deploy UBR at the Huawei App Store, where it achieves 6.6% of eCPM gain in the online A/B test and now serves the main traffic in the Huawei App Store advertising scenario.

General Purpose Recommendation System

A recommendation system is a class of information retrieval systems that provides personalized suggestions to users. Our approach is to generalize the process of creating a recommendation system so that it can be used effectively in multiple domains. This approach can be achieved by collaborative filtering by using the website’s clickstream analytics. Data related to user clicks can easily be extracted and this data's use is limited to analyse the user behaviour. We can use this extracted data to generate recommendations for the users and increase engagement on the website. The identifiers of users and items can be passed along with the user click rate to a supervised machine learning model of matrix factorization. We are using alternating least square (ALS) to predict the recommended items. The suggested system interfaces with the client system using a web API which will collect data and generate recommendations

AIM: Automatic Interaction Machine for Click-Through Rate Prediction

Feature embedding learning and feature interaction modeling are two crucial components of deep models for Click-Through Rate (CTR) prediction. Most existing deep CTR models suffer from the following three problems. First, feature interactions are either manually designed or simply enumerated. Second, all the feature interactions are modeled with an identical interaction function. Third, in most existing models, different features share the same embedding size which leads to memory inefficiency. To address these three issues mentioned above, we propose Automatic Interaction Machine (AIM) with three core components, namely, Feature Interaction Search (FIS), Interaction Function Search (IFS) and Embedding Dimension Search (EDS), to select significant feature interactions, appropriate interaction functions and necessary embedding dimensions automatically in a unified framework. Specifically, FIS component automatically identifies different orders of essential feature interactions with useless ones pruned; IFS component selects appropriate interaction functions for each individual feature interaction in a learnable way; EDS component automatically searches proper embedding size for each feature. Offline experiments on three large-scale datasets validate the superior performance of AIM. A three-week online A/B test in the recommendation service of a mainstream app market shows that AIM improves DeepFM model by 4.4% in terms of CTR.

Let’s face it: When and how facial emojis increase the persuasiveness of electronic word of mouth

Facial emojis have increasingly permeated electronic word of mouth (eWOM), but the persuasive consequences of this phenomenon remain unclear. Drawing on emotions as social information (EASI) theory, this research reveals that facial emojis influence persuasion (e.g., product choice) by affecting emotional arousal and perceived ambiguity. While the effect through emotional arousal is generally positive, the effect through ambiguity depends on the emojis’ function in eWOM: facial emojis that replace a verbal expression increase ambiguity and therefore reduce persuasion, whereas those that reiterate a verbal expression decrease ambiguity and therefore enhance persuasion. Both the emotional-arousal and ambiguity pathways determine the net persuasive effect. This research also explores two situations (high verbal context richness and eWOM from strong ties) where replacing facial emojis can increase persuasion. Finally, the authors show that facial emojis’ persuasive power is generalizable to online brand communications, influencing key management outcomes such as click-through rates for digital ads.

Survey on Click-Through Rate Prediction Based on Deep Learning

With regards to computer science, deep learning forms an essential research area. Recently, deep learning has made achievements on image processing, language understanding, data analysis, speech recognition, online advertising and so on. At the same time, display advertising has also become the most popular means of publicity with the rapid development in terms of the Internet, as well as the rapid expansion of the market of online advertising. Accurate advertising recommendation is the guarantee of Internet platform profits. The premise of accurate recommendation is accurate advertising click through rate prediction. Since 2015, the deep learning success has made the estimation of CTR results more accurate. Many CTR models have been used widely on a large amount of online platforms. This paper reviews several deep learning-based click-through rate prediction models for online advertising recently; classifies these prediction algorithms in the aspect of basic structure, complexity and main functions; and analyses the differences, advantages and the application. Finally, the survey is summarized and the future prospects of this field are envisaged.

Negativity drives online news consumption

Online media is important for society in informing and shaping opinions, hence raising the question of what drives online news consumption. Here we analyse the causal effect of negative and emotional words on news consumption using a large online dataset of viral news stories. Specifically, we conducted our analyses using a series of randomized controlled trials (N = 22,743). Our dataset comprises ~105,000 different variations of news stories from Upworthy.com that generated ∼5.7 million clicks across more than 370 million overall impressions. Although positive words were slightly more prevalent than negative words, we found that negative words in news headlines increased consumption rates (and positive words decreased consumption rates). For a headline of average length, each additional negative word increased the click-through rate by 2.3%. Our results contribute to a better understanding of why users engage with online media.

PC-IEN: a click-through rate prediction method based on dynamic collaborative personalized interest extraction

PC-IEN: a click-through rate prediction method based on dynamic collaborative personalized interest extraction

Consumers’ behavioural responses to price promotions of organic products: an introspective pre-study and an online field experiment

PurposeWhile most marketing research on organic products refers to the premium price levels of organic products, little research exists on consumers’ behavioural responses to price promotions or discounts of organic products. The present study aims to fill this research gap.Design/methodology/approachTo develop alternative hypotheses about consumers’ behavioural responses to price promotions of organic fast-moving consumer good (FMCG) products, the authors used the researcher-introspection method in a pre-study. To test the hypotheses developed based on the pre-study, the authors conducted a field experiment on online advertising of an FMCG sold in drugstores. In the field experiment, the authors exposed consumers to an online ad featuring either a price promotion (−20%) or the regular price of the product. The ads also varied in terms of whether they contained explicit organic claims or not, and whether they included implicit organic cues or not.FindingsThe price promotion increased the clickthrough rate of the ad both when combined with an explicit organic claim and when combined with the implicit cue of green product pack. The results suggest that consumers do not have significant suspicions about price promotions of organic products, but rather presume that the price promotion of an organic FMCG product is a periodical promotional action, similar to the price promotions for conventional, non-organic products. Also, consumers seem to assume that the regular prices of organic FMCG products are so high that the retailer/manufacturer can well afford periodic price discounts.Research limitations/implicationsThe present research shifts the focus of organic marketing research from the premium price levels to the effectiveness of price promotions and discounts. Further, the present results contrast with certain earlier studies that have questioned the effectiveness of price promotions for organic products.Practical implicationsThe results have different implications for marketing managers of brands not yet providing organic product versions in the market, of brands producing non-organic products, which cannot easily be rendered organic, and of brands offering organic products in the market.Originality/valueThis is, to the best of the authors’ knowledge, the first empirical study and field experiment on price promotions of organic products, including explicit organic claims.

Intelligent online selling point extraction and generation for e‐commerce recommendation

AbstractDuring the past decade, great significant advancements have been witnessed in the domain of automatic product description generation. As the services provided by e‐commerce platforms become diverse, it is necessary to dynamically adapt the patterns of descriptions generated. The selling point of products is an important type of product description for which the length should be as short as possible while still conveying key information. In addition, this kind of product description should be eye‐catching to the readers. Currently, product selling points are normally written by human experts. Thus, the creation and maintenance of these contents incur high costs. These costs can be significantly reduced if product selling points can be automatically generated by machines. In this paper, we report our experience developing and deploying the intelligent online selling point extraction (IOSPE) system to serve the recommendation system in the JD.com e‐commerce platform. In addition, an interactive tool is released for merchants in JD.COM, which additionally allow them to select and customize the selling points with flexibility. Since July 2020, IOSPE has become a core service for 62 key categories of products (covering more than four million products). So far, it has generated more than 0.1 billion selling points, thereby significantly scaling up the selling point creation operation and saving human labor. These IOSPE‐generated selling points have increased the click‐through rate (CTR) by 1.89% and the average duration the customers spent on the products by more than 2.03% compared to the previous practice, which are significant improvements for such a large‐scale e‐commerce platform.

Efficient Deep Reinforcement Learning-Enabled Recommendation

Existing recommendations based on machine learning are mainly based on supervised learning. However, these methods affected by historical behavior often bring great difficulties on mining high-quality long-tail items, achieving cold-start recommendations, and causing response inability to real-time environment changes. To this end, this paper proposes a Deep Reinforcement Learning-enabled Recommendation based on Hierarchical attention and Sample-enhanced priority experience replay (HEDRL-Rec). First, we propose a hierarchical attention mechanism to extract more hidden information, including different contributions from single feature and overall feature (comprising combined feature), for enhancing features extraction ability of Actor-Critic architecture. Then, by considering the reusability of historical experiences and differences their contributions, we then propose a sample-enhanced priority experience replay mechanism to alleviate the problems of sample imbalance, sparse data, and excessive action space, where, thereby realizing personalized recommendations in real-time changing environments. Finally, we develop a deep reinforcement learning-enabled recommendation algorithm to solve the problems of non-convergence in the Critic. Extensive experiments demonstrate that, in particular, the recommended Click-Through Rate (CTR) of the HEDRL-Rec is 10.55% higher than the state-of-the-art LIst-wise Recommendation framework based on the Deep Reinforcement learning (ILRD) scheme, while the HEDRL-Rec has better stability and usability in the recommendation scenario, effectively alleviating the cold-start problem of systems lacking manual annotation data.

Automatic product copywriting for e‐commerce

AbstractProduct copywriting is a critical component of e‐commerce recommendation platforms. It aims to attract users' interest and improve user experience by highlighting product characteristics with textual descriptions. In this paper, we report our experience deploying the proposed Automatic Product Copywriting Generation (APCG) system into the JD.com e‐commerce product recommendation platform. It consists of two main components: (1) natural language generation, which is built from a transformer–pointer network and a pretrained sequence‐to‐sequence model based on millions of training data from our in‐house platform; and (2) copywriting quality control, which is based on both automatic evaluation and human screening. For selected domains, the models are trained and updated daily with the updated training data. In addition, the model is also used as a real‐time writing assistant tool on our live broadcast platform. The APCG system has been deployed in JD.com since February 2021. By September 2021, it has generated 2.53 million product descriptions, and improved the overall averaged click‐through rate (CTR) and the conversion rate (CVR) by 4.22 and 3.61%, compared to baselines, respectively, on a year‐on‐year basis. The accumulated gross merchandise volume (GMV) made by our system is improved by 213.42%, compared to the number in February 2021.

Click-through Rate Prediction and Uncertainty Quantification Based on Bayesian Deep Learning.

Click-through rate (CTR) prediction is a research point for measuring recommendation systems and calculating AD traffic. Existing studies have proved that deep learning performs very well in prediction tasks, but most of the existing studies are based on deterministic models, and there is a big gap in capturing uncertainty. Modeling uncertainty is a major challenge when using machine learning solutions to solve real-world problems in various domains. In order to quantify the uncertainty of the model and achieve accurate and reliable prediction results. This paper designs a CTR prediction framework combining feature selection and feature interaction. In this framework, a CTR prediction model based on Bayesian deep learning is proposed to quantify the uncertainty in the prediction model. On the squeeze network and DNN parallel prediction model framework, the approximate posterior parameter distribution of the model is obtained using the Monte Carlo dropout, and obtains the integrated prediction results. Epistemic and aleatoric uncertainty are defined and adopt information entropy to calculate the sum of the two kinds of uncertainties. Epistemic uncertainty could be measured by mutual information. Experimental results show that the model proposed is superior to other models in terms of prediction performance and has the ability to quantify uncertainty.

A disaggregated interest-extraction network for click-through rate prediction

A disaggregated interest-extraction network for click-through rate prediction

High Spectral and Temporal Acuity in Primary Auditory Cortex of Awake Cats

Most accounts of single- and multi-unit responses in auditory cortex under anesthetized conditions have emphasized V-shaped frequency tuning curves and low-pass sensitivity to rates of repeated sounds. In contrast, single-unit recordings in awake marmosets also show I-shaped and O-shaped response areas having restricted tuning to frequency and (for O units) sound level. That preparation also demonstrates synchrony to moderate click rates and representation of higher click rates by spike rates of non-synchronized tonic responses, neither of which are commonly seen in anesthetized conditions. The spectral and temporal representation observed in the marmoset might reflect special adaptations of that species, might be due to single- rather than multi-unit recording, or might indicate characteristics of awake-versus-anesthetized recording conditions. We studied spectral and temporal representation in the primary auditory cortex of alert cats. We observed V-, I-, and O-shaped response areas like those demonstrated in awake marmosets. Neurons could synchronize to click trains at rates about an octave higher than is usually seen with anesthesia. Representations of click rates by rates of non-synchronized tonic responses exhibited dynamic ranges that covered the entire range of tested click rates. The observation of these spectral and temporal representations in cats demonstrates that they are not unique to primates and, indeed, might be widespread among mammalian species. Moreover, we observed no significant difference in stimulus representation between single- and multi-unit recordings. It appears that the principal factor that has hindered observations of high spectral and temporal acuity in the auditory cortex has been the use of general anesthesia.

How to Allocate White Space in Ad Design? The Impact of Product Layouts on Perceived Entitativity and Advertising Performance

White space (empty space) is an important factor in visual ad design, yet few studies focus on the specific types of white space and their influence on ad design. In this research, we systematically review white space literature; identify two types of white space, termed interspace (space in between a product layout) and circumspace (space surrounding a product layout); and investigate their effects on product layout perceptions. In four studies, we propose a not yet widely adopted practice: When holding the total white space constant, a lower interspace/circumspace ratio (more compressed product layout), compared with a higher interspace/circumspace ratio (less compressed product layout), can increase the perceived entitativity (i.e., perception of a unit as a group) of a collection of products in an ad and, ultimately, ad evaluation, purchase intentions, and click-through rates (CTRs). We further rule out potential alternative explanations and identify a theoretically and practically relevant boundary condition: If the products are framed as conceptually dissimilar, the perception of entitativity becomes less important, which reduces ad performance. We conclude by discussing the theoretical and practical implications of our research.

Difference Between Screw Cement Filling and Adequate Osteotomy With Thick Liner for Primary Total Knee Arthroplasty in Patients With Rand IIb Tibial Defects

BackgroundThe effectiveness of 2 treatment options, screw-cement fill, and adequate osteotomy with a thick liner, in treating patients with Rand IIb tibial defects (tibial plateau defects to a depth of 5 to 10 millimeters) in primary total knee arthroplasty (TKA) has not yet been demonstrated. Therefore, we performed a retrospective study to evaluate the differences between these 2 treatments. MethodsWe retrospectively analyzed patients who underwent primary TKA for Rand IIb tibial plateau defects from 2015 to 2020 from a department database. Patients were categorized into the screw-cement and thick liner groups based on the different options used to repair tibial defects. We evaluated Knee Society Score, range of motion (ROM), Insall-Salvati index (ISI), and Forgotten Joint Score (FJS) in both groups. We also compared differences in prosthesis survival, stiffness, myasthenia, and joint clicking between the 2 groups at mean 2 years postoperatively (range, 2 to 2.3). A power analysis was performed on the number of cases in the cohort. ResultsPostoperative femur-tibia mechanical axis (FTMA) correction was significantly higher in the screw-cement group than in the thick-liner group: 18.8 (±5.6°) versus 15.4 (±5.9°) (P < .01); At mean 2 years after surgery, the American Knee Society Functional Score improvement values were higher in the thick-liner group than in the screw-cement group: 36.3 (±12.4) versus 42.4 (±16.4) (P = .05). Postoperative ISI scores were 0.95 (±0.12) points in the screw-cement group and 0.89 (±0.13) points in the-thick liner group (P = .03). There were no statistically significant differences in the Knee Society Clinical Score, ROM, FJS, stiffness, myasthenia, joint clicking, and revision rate. ConclusionThe results of this study showed no significant difference in clinical outcomes between the 2 reconstruction strategies of the screw cement fill technique and the adequate osteotomy and thick liner technique for Rand IIb tibial plateau defects. However, in patients who have FTMA deformities greater than 20° or in younger patients who need to preserve bone volume, we recommend the screw cement filling technique to ensure stable postoperative results and to prepare these patients for possible later surgery.

Knowledge distillation-enhanced multitask framework for recommendation

With the ever-growing amount of online information, recommender systems (RSs) act as information filtering tools and are widely used in various e-commerce platforms. Recommender methods generally adopt only one type of behavior data for single-task learning. Multitask learning is commonly used to simultaneously address multiple tasks utilizing as much information as possible. We propose a knowledge distillation-enhanced shared-bottom model for recommender multitask learning. The postview click-through rate, postview conversion rate and average transaction value are learned together to determine the final gross merchandise volume. First, a mixture of experts with gate networks is used as a shared bottom to learn task-specific representations for all tasks, while a tree distillation framework is designed as an expert for better feature selection. Then, we design a neural factorization machine as a task-specific prediction network to estimate individual goals. Finally, an objective function is designed based on task outputs to optimize the gross merchandise volume. Our proposed framework can enhance recommender preferences while avoiding generating a large number of parameters.This model serves the e-commerce scenario, with the following specific use scenario: building personalized GMV predictions for small scenarios on e-commerce platforms at low cost, which is a 0 to 1 process that is often highly rewarding. Ultimately, we propose to improve the reliability of e-commerce scenarios.

RESUS: Warm-up Cold Users via Meta-learning Residual User Preferences in CTR Prediction

Click-through Rate (CTR) prediction on cold users is a challenging task in recommender systems. Recent researches have resorted to meta-learning to tackle the cold-user challenge, which either perform few-shot user representation learning or adopt optimization-based meta-learning. However, existing methods suffer from information loss or inefficient optimization process, and they fail to explicitly model global user preference knowledge, which is crucial to complement the sparse and insufficient preference information of cold users. In this article, we propose a novel and efficient approach named RESUS, which decouples the learning of global preference knowledge contributed by collective users from the learning of residual preferences for individual users. Specifically, we employ a shared predictor to infer basis user preferences, which acquires global preference knowledge from the interactions of different users. Meanwhile, we develop two efficient algorithms based on the nearest neighbor and ridge regression predictors, which infer residual user preferences via learning quickly from a few user-specific interactions. Extensive experiments on three public datasets demonstrate that our RESUS approach is efficient and effective in improving CTR prediction accuracy on cold users, compared with various state-of-the-art methods.

Serving two masters? Optimizing mobile ad contracts with heterogeneous advertisers

Our research question is motivated by a real problem faced by an existing demand aggregator. The aggregator represents multiple advertisers, each of whom signs one of two types of contracts with the aggregator for bidding on an RTB (real‐time bidding) platform. A quality contract occurs on a cost‐per‐impression (CPM, i.e., cost per thousand impressions) basis. The advertiser is promised a minimum number of impressions and pays a CPM that is a function of the targeting quality as measured by the realized click‐through rate (CTR). In a performance contract the advertiser pays on a CPC (cost‐per‐click) basis constrained by a budget. We develop and solve a generalized profit maximization problem that jointly optimizes the aggregator's bidding and allocation decisions. The allocation policy optimally assigns each arriving bidding opportunity to a specific advertiser. The bidding policy then computes a bid amount for that allocation based on the estimated click probability of the opportunity. Our solution has nice theoretical properties. First, neither policy depends on the memory carried in the system, that is, the sequence of previous states and decisions, making the solution easy to implement. Second, the allocation policy is shown to have a threshold structure. This enables the assignment of arriving opportunities into one of two distinct sets, each corresponding to a specific advertising contract type. The assessment of the impact of the change in various parameters on the solution is used to derive several interesting and important implications for the management of advertising contracts.

Click-through rate prediction model integrating user interest and multi-head attention mechanism

The purpose of click-through rate (CTR) prediction is to anticipate how likely a person is to click on an advertisement or item. It's required for a lot of internet applications, such online advertising and recommendation systems. The previous click-through rate estimation approach suffered from the following two flaws. On the one hand, input characteristics (such as user id, user age, user age, item id, item category) are usually sparse and multidimensional, making them effective. High-level combination characteristics are used for prediction. Obtaining it manually by domain experts takes a long time and is difficult to finish; also, customer interests are not all the same. The accuracy of the model findings will significantly increase if this immediately recognized component is incorporated in the prediction model. As a consequence, this study creates an IARM (interactive attention rate estimation model) that incorporates user interest as well as a multi-head self-attention mechanism. The deep learning network is used in the model to determine the user's interest expression based on user attributes. The multi-head self-attention mechanism with residual network is then employed to get feature interaction, which enhances the degree of effect of significant characteristics on the estimation result as well as its accuracy. The IARM model outperforms other recent prediction models in the assessment metrics AUC and LOSS, and it has superior accuracy, according to the results from the public experimental data set.