The FinRec workshop series offers a central forum for the study and discussion of the domain-specific aspects, challenges, and opportunities of RecSys and other related technologies in the financial services domain. Six years after the second edition of the workshop, the recent advances in the area of personalization and recommendation in financial services fostered the need for a new workshop aiming at bringing together researchers and practitioners working in financial services-related areas. Accordingly, the third edition of the event aims to: (1) understand and discuss open research challenges, (2) provide an overview of existing technologies using recommender systems in the financial services domain, and (3) provide an interactive platform for information exchange between industry and academia.

During the past decade, recommender systems have rapidly become an indispensable element of websites, apps, and other platforms that seek to provide personalized interactions to their users. As recommendation technologies are applied to an ever-growing array of non-standard problems and scenarios, researchers and practitioners are also increasingly faced with challenges of dealing with greater variety and complexity in the inputs to those recommender systems. For example, there has been more reliance on fine-grained user signals as inputs rather than simple ratings or likes. Applications require more complex domain-specific constraints on inputs to the recommender systems. Likewise, the outputs of recommender systems are moving towards more complex composite items, such as package or sequence recommendations. This increasing complexity requires smarter recommender algorithms that can deal with this diversity in inputs and outputs. For the past four years, the ComplexRec workshop series has offered an interactive venue for discussing approaches to recommendation in complex scenarios that have no simple one-size-fits-all solution.

In this report, we offer a brief overview of the contributions and takeaways from the Joint KaRS & ComplexRec Workshop, co-located with the 15th edition of the ACM RecSys in Amsterdam, The Netherlands. With this workshop, we aimed to merge the main objectives envisioned for the 3rd Edition of the Workshop of Knowledge-aware and Conversational Recommender Systems and the 5th Edition of the Workshop on Recommendation in Complex Environments. This joint workshop adopted a hybrid format aligned with the goal of this year's main conference congregating to continue to build community around recommender systems research and development. Date: 27th September-1st October, 2021. Website: https://kars-workshop.github.io/2021/program/.

During the past decade, recommender systems have rapidly become an indispensable element of websites, apps, and other platforms that are looking to provide personalized interaction to their users. As recommendation technologies are applied to an ever-growing array of non-standard problems and scenarios, researchers and practitioners are also increasingly faced with challenges of dealing with greater variety and complexity in the inputs to those recommender systems. For example, there has been more reliance on fine-grained user signals as inputs rather than simple ratings or likes. Many applications also require more complex domain-specific constraints on inputs to the recommender systems. The outputs of recommender systems are also moving towards more complex composite items, such as package or sequence recommendations. This increasing complexity requires smarter recommender algorithms that can deal with this diversity in inputs and outputs. The ComplexRec workshop series offers an interactive venue for discussing approaches to recommendation in complex scenarios that have no simple one-size-fits-all solution.

During the past decade, recommender systems have rapidly become an indispensable element of websites, apps, and other platforms that are looking to provide personalized interaction to their users. As recommendation technologies are applied to an ever-growing array of non-standard problems and scenarios, researchers and practitioners are also increasingly faced with challenges of dealing with greater variety and complexity in the inputs to those recommender systems. For example, there has been more reliance on fine-grained user signals as inputs rather than simple ratings or likes. Many applications also require more complex domain-specific constraints on inputs to the recommender systems. The outputs of recommender systems are also moving towards more complex composite items, such as package or sequence recommendations. This increasing complexity requires smarter recommender algorithms that can deal with this diversity in inputs and outputs. The ComplexRec workshop series offers an interactive venue for discussing approaches to recommendation in complex scenarios that have no simple one-size-fits-all solution.

Several tasks in information retrieval (IR) rely on assumptions regarding the distribution of some property (such as term frequency) in the data being processed. This thesis argues that such distributional assumptions can lead to incorrect conclusions and proposes a statistically principled method for determining the "true" distribution. This thesis further applies this method to derive a new family of ranking models that adapt their computations to the statistics of the data being processed. Experimental evaluation shows results on par or better than multiple strong baselines on several TREC collections. Overall, this thesis concludes that distributional assumptions can be replaced with an effective, efficient and principled method for determining the "true" distribution and that using the "true" distribution can lead to improved retrieval performance.

Over the past decade, recommendation algorithms for ratings prediction and item ranking have steadily matured. However, these state-of-the-art algorithms are typically applied in relatively straightforward scenarios. In reality, recommendation is often a more complex problem: it is usually just a single step in the user's more complex background need. These background needs can often place a variety of constraints on which recommendations are interesting to the user and when they are appropriate. However, relatively little research has been done on these complex recommendation scenarios. The ComplexRec 2018 workshop addresses this by providing an interactive venue for discussing approaches to recommendation in complex scenarios that have no simple one-size-fits-all solution.

What if Information Retrieval (IR) systems did not just retrieve relevant information that is stored in their indices, but could also" understand" it and synthesise it into a single document? We present a preliminary study that makes a first step towards answering this question. Given a query, we train a Recurrent Neural Network (RNN) on existing relevant information to that query. We then use the RNN to" deep learn" a single, synthetic, and we assume, relevant document for that query. We design a crowdsourcing experiment to assess how relevant the" deep learned" document is, compared to existing relevant documents. Users are shown a query and four wordclouds (of three existing relevant documents and our deep learned synthetic document). The synthetic document is ranked on average most relevant of all.

A labeling scheme is a method of distributing the information about the structure of a graph among its vertices by assigning short labels, such that a selected function on pairs of vertices can be computed using only their labels. A labeling scheme consists of an encoder that has access to the entire graph and assigns labels to vertices, and a decoder that has access to only the labels of a smaller set of vertices (typically a pair) and returns information about this subset (e.g., whether two vertices are adjacent, or the distance between them in the graph). The main objective is to minimize the maximum label size: the maximum number of bits used in a label of any vertex. Among the applications of labeling schemes are XML search engines, mapping services, and internet routing.

What if Information Retrieval (IR) systems did not just retrieve relevant information that is stored in their indices, but could also "understand" it and synthesise it into a single document? We present a preliminary study that makes a first step towards answering this question. Given a query, we train a Recurrent Neural Network (RNN) on existing relevant information to that query. We then use the RNN to "deep learn" a single, synthetic, and we assume, relevant document for that query. We design a crowdsourcing experiment to assess how relevant the "deep learned" document is, compared to existing relevant documents. Users are shown a query and four wordclouds (of three existing relevant documents and our deep learned synthetic document). The synthetic document is ranked on average most relevant of all.

Several properties of information retrieval (IR) data, such as query frequency or document length, are widely considered to be approximately distributed as a power law. This common assumption aims to focus on specific characteristics of the empirical probability distribution of such data (e.g., its scale-free nature or its long/fat tail). This assumption, however, may not be always true. Motivated by recent work in the statistical treatment of power law claims, we investigate two research questions: (i) To what extent do power law approximations hold for term frequency, document length, query frequency, query length, citation frequency, and syntactic unigram frequency? And (ii) what is the computational cost of replacing ad hoc power law approximations with more accurate distribution fitting? We study 23 TREC and 5 non-TREC datasets and compare the fit of power laws to 15 other standard probability distributions. We find that query frequency and 5 out of 24 term frequency distributions are best approximated by a power law. All remaining properties are better approximated by the Inverse Gaussian, Generalized Extreme Value, Negative Binomial, or Yule distribution. We also find the overhead of replacing power law approximations by more informed distribution fitting to be negligible, with potential gains to IR tasks like index compression or test collection generation for IR evaluation.

Caching posting lists can reduce the amount of disk I/O required to evaluate a query. Current methods use optimisation procedures for maximising the cache hit ratio. A recent method selects posting lists for static caching in a greedy manner and obtains higher hit rates than standard cache eviction policies such as LRU and LFU. However, a greedy method does not formally guarantee an optimal solution. We investigate whether the use of methods guaranteed, in theory, to find an approximately optimal solution would yield higher hit rates. Thus, we cast the selection of posting lists for caching as an integer linear programming problem and perform a series of experiments using heuristics from combinatorial optimisation (CCO) to find optimal solutions. Using simulated query logs we find that CCO yields comparable results to a greedy baseline using cache sizes between 200 and 1000 MB, with modest improvements for queries of length two to three.

Document coherence describes how much sense text makes in terms of its logical organisation and discourse flow. Even though coherence is a relatively difficult notion to quantify precisely, it can be approximated automatically. This type of coherence modelling is not only interesting in itself, but also useful for a number of other text processing tasks, including Information Retrieval (IR), where adjusting the ranking of documents according to both their relevance and their coherence has been shown to increase retrieval effectiveness [34,37]. The state of the art in unsupervised coherence modelling represents documents as bipartite graphs of sentences and discourse entities, and then projects these bipartite graphs into one-mode undirected graphs. However, one-mode projections may incur significant loss of the information present in the original bipartite structure. To address this we present three novel graph metrics that compute document coherence on the original bipartite graph of sentences and entities. Evaluation on standard settings shows that: (i) one of our coherence metrics beats the state of the art in terms of coherence accuracy; and (ii) all three of our coherence metrics improve retrieval effectiveness because, as closer analysis reveals, they capture aspects of document quality that go undetected by both keyword-based standard ranking and by spam filtering. This work contributes document coherence metrics that are theoretically principled, parameter-free, and useful to IR.

Divergence From Randomness (DFR) ranking models assume that informative terms are distributed in a corpus differently than non-informative terms. Different statistical models (e.g. Poisson, geometric) are used to model the distribution of non-informative terms, producing different DFR models. An informative term is then detected by measuring the divergence of its distribution from the distribution of non-informative terms. However, there is little empirical evidence that the distributions of non-informative terms used in DFR actually fit current datasets. Practically this risks providing a poor separation between informative and non-informative terms, thus compromising the discriminative power of the ranking model. We present a novel extension to DFR, which first detects the best-fitting distribution of non-informative terms in a collection, and then adapts the ranking computation to this best-fitting distribution. We call this model Adaptive Distributional Ranking (ADR) because it adapts the ranking to the statistics of the specific dataset being processed each time. Experiments on TREC data show ADR to outperform DFR models (and their extensions) and be comparable in performance to a query likelihood language model (LM).

An adjacency labeling scheme is a method that assigns labels to the vertices of a graph such that adjacency between vertices can be inferred directly from the assigned label, without using a centralized data structure. We devise adjacency labeling schemes for the family of power-law graphs. This family that has been used to model many types of networks, e.g. the Internet AS-level graph. Furthermore, we prove an almost matching lower bound for this family. We also provide an asymptotically near- optimal labeling scheme for sparse graphs. Finally, we validate the efficiency of our labeling scheme by an experimental evaluation using both synthetic data and real-world networks of up to hundreds of thousands of vertices.

We present two novel models of document coherence and their application to information retrieval (IR). Both models approximate document coherence using discourse entities, e.g. the subject or object of a sentence. Our first model views text as a Markov process generating sequences of discourse entities (entity n -grams); we use the entropy of these entity n -grams to approximate the rate at which new information appears in text, reasoning that as more new words appear, the topic increasingly drifts and text coherence decreases. Our second model extends the work of Guinaudeau & Strube [28] that represents text as a graph of discourse entities, linked by different relations, such as their distance or adjacency in text. We use several graph topology metrics to approximate different aspects of the discourse flow that can indicate coherence, such as the average clustering or betweenness of discourse entities in text. Experiments with several instantiations of these models show that: (i) our models perform on a par with two other well-known models of text coherence even without any parameter tuning, and (ii) reranking retrieval results according to their coherence scores gives notable performance gains, confirming a relation between document coherence and relevance. This work contributes two novel models of document coherence, the application of which to IR complements recent work in the integration of document cohesiveness or comprehensibility to ranking [5, 56].

Typically search engine results (SERs) are presented in a ranked list of decreasing estimated relevance to user queries. While familiar to users, ranked lists do not show inherent connections between SERs, e.g. whether SERs are hyperlinked or authored by the same source. Such potentially useful connections between SERs can be displayed as graphs. We present a preliminary comparative study of ranked lists vs graph visualisations of SERs. Experiments with TREC web search data and a small user study of 10 participants show that ranked lists result in more precise and also faster search sessions than graph visualisations.

One of the greatest challenges in product development is to build a product that customers value more than their next-best alternative. Accomplishing this task requires a shared understanding between marketing and engineering of the targeted customers and their careabouts. Unfortunately, in many businesses, marketing and engineering struggle to effectively represent the customer in the product development process. This paper describes Texas Instruments' use of conjoint analysis and design of experiments modelling in the development of projection display products.