Creative influence is responsible for a considerable part of the creative process of an artist and can largely be associated with their social circle. It has been observed that the type and amount of relationships with other fellow artists correlates with the success of an artist. Most of the recent literature has focused on using artefact similarity as a proxy for creative influence between two artists. However, this approach neglects the significance of an artist's social network or flattens the individuality of a relationship by only addressing it as a direct connection. In this work, we propose an ontology to comprehensively model the relationship between individuals as a Knowledge Graph. Additionally, we design and implement an explainable method based on graph theory to predict the influences of an artist given their social network. We evaluate our method on a dataset of relationships between Jazz musicians and achieve accurate results when compared to baselines that rely on the distribution of the data. Our results are aligned with relevant works from the socio-cognitive and psychology fields. 1
Identifying, capturing and hypothesising the influences of an artist is an important aspect that
an art critic considers when analysing an artefact or, in general, the artist themselves [
Yet, the importance of capturing and understanding the influences of an artist has a great
impact on many pragmatic aspects. Mitali and Ingram [
CREAI 2023 - Workshop on Artificial Intelligence and Creativity, November 6-9, 2023, Rome, Italy
those relationships, the more an artist gets deeper into a clique formed by other meaningful
artists, the more its work is acclaimed by critics. For instance, the success of the band The Velvet
Underground is often partially associated to their relationship with the artist Andy Warhol [
To understand the influence on an artist, it is important to take into account its social
relationships as well. Most recent approaches, however, use artefact similarity as a proxy for
creative influence Abe et al. [
In this work, we propose a method to predict the influence of musical artists by only taking
into account their social network. We design an ontology to model the relationships between
artists in an expressive way. Rather than consider them as simple direct relationships we
model them as complex situations that involve diferent agents and concepts. We refactor the
data from the Linked Jazz project [
Our contributions can be summarised as follows: • an ontology to model the relationships between human agents, with a particular focus on artists; • an explainable method to predict creative influence between artists. 1Retrieved from https://triplydb.com/pratt/linked-jazz/
The paper is structured as follows: in Section 2 we provide a review of related works addressing the prediction and identification of influence between artists. In Section 3 we present the ontology and its associated method to predict creative influence between artists. In Section 4 we describe the experimental setup while in Section 5 we present the obtained results. Section 6 summarises the results of previous sections and highlights potential extensions and future work.
Evaluating the influences of an artist, and in particular of a composer or a musician, is mostly considered a subjective task. Usually, experts analyse the compositions of an artist in a critical way to relate them to other important artists. One approach to detecting creative influence is to directly analyse explicit influence connections, curated by human annotators. Smith and Georges [16], for example, analyses the influences identified in the Classical Music Navigator (CMS) to better understand the influence of the composers in the dataset. A similar approach is taken by Georges and Seckin [17], where the data on creative influence is used to investigate the similarity of musical compositions.
Relying on similarity as a proxy for creative influence is a popular approach that has been
explored using diferent techniques. Abe et al. [
Relying on artefact similarity, however, can be a problematic approach in art. Influence can
afect an artist in a negative way, in the sense that the influenced artist deliberately abstains
from his influence [
Diferently from the described approaches, we investigate the importance of social relationships without taking into account any information on the creative artefact. Our approach can be easily integrated with other methods that use similarity, to yield a more general method for uncovering hidden relations when perceptual similarity is the only measure taken into account.
This section provides a detailed description of our method. In Section 3.1 we discuss the design and implementation of the ontology. In Section 3.2 we describe in detail the algorithm used to compute the influences between entities in the KG. In Section 3.3 we describe the procedure designed to learn the weight of each relationship type.
The ontology is built upon the concept of social relation from the DOLCE ontology [26]. A social relation can be defined as a situation expressed by some source of information 2, some participants and a role that qualifies the type of the relation.
In our ontology, we only take into account pairwise relationships. This is intended as two roles (source and target) that partake in the relationship. In this way, it is possible to define a relation between sets of entities while retaining the directedness of the relationship. An example is the Mentorship relation, where a single mentor might have multiple students. The source of information tracks the provenance of the relationship assertion. The relationship types are described in Table 1.
Figure 1 visually represents the ontology. The class :PersonalRelationship reifies the relationship between two agents. To guarantee compactness and easiness of querying, we define the property :hasPersonalRelationshipWith to instantiate the binary projection reified by :PersonalRelationship. This is defined through the use of property chain axioms (blue box in Figure 1). Note that we do not restrict the amount of subjects to the :hasSource and :hasTarget properties. This enables the representation of pairwise relationships between two sets of entities.
In Figure 2 an example of how the ontology can be used to define the friendship relationship of Table 1 is described. In order to define a friendship relationship between the musicians Trent Reznor and David Bowie it is suficient to add the triple < Trent Reznor, :hasFriend, David Bowie > in the Knowledge Graph. The reification of the relationship is automatically performed by the inference engine. 2An information object in DOLCE dul: http://www.ontologydesignpatterns.org/ont/dul/DUL.owl#
: https://relationship-ontology/ :isSourceOf
:hasTarget ⊑ :hasPersonalRelationshipWith owl:Thing
dul:SocialRelation dul:isExpressedBy rdf:subClassOf
:hasPersonal RelationshipWith
dul:Agent :hasSource :hasTarget David Bowie :hasFriend :hasPersonal RelationshipWith
Trent Reznor :PersonalRelationship dul:isClassifiedBy :RelationshipType
The implemented reification allows us to represent the relationship as a whole rather than lfattening it into a binary relation, resulting in a richer characterisation of the relationship and a high degree of control in further refining it. For example, in Figure 2 the dashed properties represent refinement operations over the initial definition. It is possible to classify the relationship as both a friendship and mentorship relationship while adding documents that act as references to back up the assertion.
Once relations are represented using the ontology described in Section 3.1, the resulting Knowledge Graph can be interpreted as a semantically defined social network. By only relying on the binary projections of the reified relationships we can extract a directed graph where entities are directly connected to each other by means of a set of edges . In order to quantify the influence of one artist on another, we exploit tools that pertain to the analysis of complex networks.
Our approach is based on the -communicability [
has friend has bandmate admires
Trent Reznor has mentor has mentor
David Bowie Pete Townshend has mentor
Brian Eno
Halsey
In Figure 3 a visual example of the communicability between artists and the importance of the length of the path is reported.
We rely on -communicability as an indicator of how influential an artist is with respect to the other artists in the Knowledge Graph. We define a Knowledge Graph as a directed edge-labelled graph = (, , ) where represents the set of nodes in the graph, the set of edges and the set of labels that can be assigned to an edge ∈ . is efectively the set of binary projections obtained from the relationship types of Table 1.
The -communicability between nodes , in a graph is computed as () where is a suitable matrix function and is the adjacency matrix of . In order to take into account the + diferent importance of the relationship types ∈ we use a weighting function : → R0 . The function can be interpreted as the absolute degree of importance of a relationship type ⎜⎜⎝⎜ 010 0 and can either be predefined or learned, as we show in section 3.3. We can now define the weighted adjacency matrix as (1) (2) (3) where (, ) is the set of labels of the edges between and .
The -communicability (, ) between the nodes , ∈ is computed as =
∑︁ ∈(,)
() ∞ (, ) = ∑︁ ()
=1 where is the -th power of , and is a weight assigned to the walks of length . Given an adjacency matrix of a graph , the entry () is equal to the sum of the weights of all walks in from node to node of length exactly [28].
The weight defines how the importance of a walk should decay as a function of its length
and needs to be carefully chosen in order to make sure that the sum of Equation 2 converges
to the finite value. This is done by using a succession ( ) converging to 0 [
! =1 where is a parameter corresponding to the maximum length of a walk taken into account by . An approximation of the centrality of a node ∈ can be obtained by computing (, ). See Figure 4 for an example on how Equation 3 is computed based on Figure 3.
One requirement of the method just described is that the graph from which the influence is predicted is weighted.
Defining the weight of a social relationship is an elaborate task, particularly when it needs to
be contextualised in the creative domain. Perry-Smith [23] argues on the existence of strong
and weak social ties and their influence on creativity. Strong ties, defined as highly redundant
connections between two individuals (e.g. Trent Reznor and David Bowie in Figure 3) are found to
positively correlate with creativity. A straightforward approach, which will serve as a baseline,
is to assign to each relation type the same weight (e.g. 1) by using : → 1. However, it
is important to note that not all relationships equally correlate with creativity [
We propose to learn the weights assigned by by fitting the data in the knowledge graph. This can be obtained by framing the problem of predicting creative influence as a multi-class classification problem. Given an edge = (, ) we can interpret (, ) as the probability that ∈ with ∈ () where is the label assigned to the edges that semantically states that is creatively influenced by . Essentially, the -communicability of a pair of nodes (, ) measures the degree to which is creatively influenced by .
In order to do that we minimise the cross entropy-based learning-to-rank loss defined in Bruch [31] between (^ ) and where is the label assigned to the relationship that represents the influence of an artist onto another artist and ^ = ∑︀ . By relying on ∈ a learning-to-rank loss we learn weights in such a way that, given an input artist, its most influential artists are given a high weight. As a result, the model will be able to rank other artists in a meaningful way despite the absence of any explicit edge in the original graph.
While aggregating relations as in Equation 1 is a natural and straightforward approach, it is reasonable to suppose that the joint presence of two relationships, e.g. friendship and mentorship together, might be more (or less) important than the sum of the two components. To take this additional consideration into account we perform a non-linear combination using a feedforward neural network with one hidden layer using a ReLU activation. Equation 1 is hence updated to = ([ ∀ ∈ ]) (4) with being the function learned by the neural network.
In this section, we describe the experiment that we perform on the method proposed in Section 3. In Section 4.1 we describe the creation of the Knowledge Graph that is used to estimate and evaluate the predicted creative influence while in Section 4.2 we describe the experimental setup used to assess the accuracy of the method from Section 3.
We rely on the data from the Linked Jazz project [
We align the Linked Jazz KG to our ontology (described in Section 3.1) through the use of a SPARQL construct query.
The KG contains a total of 5058 statements between 70 artists, where each musician has 72 relations asserted on average. In Figure 5 the distribution of the relationship between the entities in the KG is shown.
We experiment with the methods of Section 3 on the Knowledge Graph described in the previous section.
In order to learn the weights of the function of Equation 1 we split the data into the usual training and testing partitions, where the testing partition is 20% of the total data. The resulting training and testing data are hence composed of, respectively, 56 and 14 artists. Since the amount of explicitly stated influences available in the KG is much lower than the total amount of edges, the split is only based on nodes that have some influence edges asserted. This allows us to efectively evaluate the accuracy of the model with respect to creative influence prediction. Given the small size of both splits, we evaluate the models on aggregate values from a set of 5 distinct experiments on diferent subsets of the data. This helps us mitigate the noise due to the low amount of data available.
We evaluate each model using Mean Reciprocal Rank (MRR), Mean Average Precision (MAP) and Discounted Cumulative Gain (DCG). All the listed metrics measure how high are ranked appropriate values, e.g. how high are ranked actual influential artists with respect to a reference artist. MRR can be interpreted as how far is the first influential artist in the ordered list. MAP is the average of the number of relevant entries within the first results, where is the number of influences of each artist. DCG evaluates the results by penalising when relevant entries are not positioned at the top of the list. Each model is trained using COCOB [32], a parameter-free optimisation method. All the experiments are performed on an Intel i9 with 128 of RAM and an Nvidia RTX3090 with 24 of VRAM.
In Figure 6 results from all methods are aggregated and plotted as a function of the degree of Equation 2. The results highlight how using walks whose length extends at most up to 2 nodes obtains the best results. Influences from artists that are dificult to reach, i.e. that are separated by many other nodes, add noise to the -communicability matrix. In fact, Perry-Smith [23] argues that creative influence can be classified into two main categories: strong and weak influence, where strong influence, as opposed to weak influence, is the result of many redundant connections between two nodes. In other words, the influence of an artist on another artist is stronger if the amount of connections between the two is high. Taking into account long walks results in many connections that eventually turn a weak influence into a strong one. This can also be seen in the example of Figure 2. The influence of Pete Townshend on Halsey can be safely ignored. Taking into account walks longer than degree 2 wrongly detects this relationship as a strong one. In Equation 2 longer walks are considered less important. However, this proves not to be enough, as the number of distinct longer walks from two entities mitigates this discount and results in less accurate predictions. Further investigation on other converging series used for decaying weights can result in more accurate performances.
Table 2 reports the results obtained from the experiments of Section 4 with = 2. Learning the function used in Equation 3 leads to the best results on aggregate with respect to all the metrics taken into account. Surprisingly, using a neural network as illustrated in Equation 4 does not result in a definite gain with respect to the simpler model of Equation 1. The main reason for that is the lack of training data. The network is not able to generalise over the target task and tends to overfit in the training data despite the small number of parameters.
In Table 3 the weights learned by the best method of Table 2 are described. Judging from the
mean and median values, the most important relationships are the friendship and mentorship
relationship. The former aligns with the findings of [
In this work, we present a novel method, described in Section 3, to detect creative influence
between artists using techniques based on graph theory and complex network science. Our
method takes into account the individuality of a relationship type through the use of an ontology
illustrated in Section 3.1. By framing the influence prediction task as a classification task we
are able to obtain an interpretable model that performs better than robust baselines. The
results described in Section 5 highlight how a straightforward combination of the diferent
graph planes identified by the diferent relationship types results in accurate results. Moreover,
the weights assigned to each relationship type are in line with relevant socio-cognitive and
psychological findings, thus additionally validating the results. Our attempt to increase the
accuracy of our results through the use of a machine learning approach led to less accurate
predictions. Nonetheless, it is dificult to objectively rule out the possibility of combining
machine learning techniques with our methods. In future works, we plan on extending the
dataset available. The main problem with the experiments relying on the neural network can
indeed be partially caused by the limited amount of training data for the model. An approach is
to employ data augmentation techniques, in order to exploit the data as much as possible and
reduce the chances of overfitting the model. With the availability of additional data, for instance
by using relation extraction methods [33, 34, 35], more complex architectures can also be used,
such as attention-based models [36]. Finally, we would like to explore clustering methods and
detect communities of artists within the -communicability matrix obtained from the method
of Section 3.2. This would enable the identification of cliques of artists that are socially related
and hence provide a tool to better understand the creative process of an artist. Combining the
relationship types with additional relevant information, such as the socio-cultural context [20],
is also an interesting improvement worth investigating. Similarly, combining our approach with
the one identified by Saleh et al. [
The authors would like to thank Daniele Marini for his exploratory work on the subject. This project has received funding from the FAIR – Future Artificial Intelligence Research foundation as part of the grant agreement MUR n. 341. functions, SIAM J. Matrix Anal. Appl. 39 (2018) 1521–1546. URL: https://doi.org/10.1137/ 17M1133920. doi:10.1137/17M1133920. [16] C. H. Smith, P. Georges, Composer similarities through “the classical music navigator”: Similarity inference from composer influences, Empirical Studies of the Arts 32 (2014) 205–229. [17] P. Georges, A. Seckin, Music information visualization and classical composers discovery: an application of network graphs, multidimensional scaling, and support vector machines, Scientometrics 127 (2022) 2277–2311. URL: https://doi.org/10.1007/s11192-022-04331-8. doi:10.1007/s11192-022-04331-8. [18] A. E. White, J. C. Kaufman, M. Riggs, How “outsider” do we like our art?: Influence of artist background on perceptions of warmth, creativity, and likeability., Psychology of Aesthetics, Creativity, and the Arts 8 (2014) 144. [19] E. Stamkou, G. A. van Kleef, A. C. Homan, The art of influence: When and why deviant artists gain impact., Journal of Personality and Social Psychology 115 (2018) 276. [20] M. Mauskapf, E. Quintane, N. Askin, J. M. Mol, Embeddedness and the production of novelty in music: A multi-dimensional perspective, Academy of Management Proceedings 2017 (2017) 16678. URL: https://doi.org/10.5465/AMBPP.2017.2. doi:10.5465/AMBPP.2017.2. arXiv:https://doi.org/10.5465/AMBPP.2017.2. [21] K. J. Borowiecki, Good reverberations? teacher influence in music composition since 1450,
Journal of Political Economy 130 (2022) 991–1090. [22] J. E. Perry-Smith, C. E. Shalley, The social side of creativity: A static and dynamic social network perspective, Academy of management review 28 (2003) 89–106. [23] J. E. Perry-Smith, Social yet creative: The role of social relationships in facilitating individual creativity, Academy of Management journal 49 (2006) 85–101. [24] M. C. Pattuelli, K. Hwang, M. Miller, Accidental discovery, intentional inquiry: Leveraging linked data to uncover the women of jazz, Digit. Scholarsh. Humanit. 32 (2017) 918–924.
URL: https://doi.org/10.1093/llc/fqw047. doi:10.1093/llc/fqw047. [25] J. Lethem, The ecstasy of influence (2007). [26] S. Borgo, R. Ferrario, A. Gangemi, N. Guarino, C. Masolo, D. Porello, E. M. Sanfilippo, L. Vieu, DOLCE: A descriptive ontology for linguistic and cognitive engineering, Appl. Ontology 17 (2022) 45–69. URL: https://doi.org/10.3233/AO-210259. doi:10.3233/ AO-210259. [27] A. Krisnadhi, F. Maier, P. Hitzler, OWL and rules, in: A. Polleres, C. d’Amato, M. Arenas, S. Handschuh, P. Kroner, S. Ossowski, P. F. Patel-Schneider (Eds.), Reasoning Web. Semantic Technologies for the Web of Data - 7th International Summer School 2011, Galway, Ireland, August 23-27, 2011, Tutorial Lectures, volume 6848 of Lecture Notes in Computer Science, Springer, 2011, pp. 382–415. URL: https://doi.org/10.1007/978-3-642-23032-5_7. doi:10. 1007/978-3-642-23032-5\_7. [28] J. A. Bondy, U. S. R. Murty, Graph Theory, Graduate Texts in Mathematics, Springer, 2008.
URL: https://doi.org/10.1007/978-1-84628-970-5. doi:10.1007/978-1-84628-970-5. [29] L. Katz, A new status index derived from sociometric analysis, Psychometrika 18 (1953) 39–43. [30] E. Estrada, N. Hatano, Communicability graph and community structures in complex networks, Appl. Math. Comput. 214 (2009) 500–511. URL: https://doi.org/10.1016/j.amc. 2009.04.024. doi:10.1016/j.amc.2009.04.024. [31] S. Bruch, An alternative cross entropy loss for learning-to-rank, in: J. Leskovec, M. Grobelnik, M. Najork, J. Tang, L. Zia (Eds.), WWW ’21: The Web Conference 2021, Virtual Event / Ljubljana, Slovenia, April 19-23, 2021, ACM / IW3C2, 2021, pp. 118–126. URL: https://doi.org/10.1145/3442381.3449794. doi:10.1145/3442381.3449794. [32] F. Orabona, T. Tommasi, Training deep networks without learning rates through coin betting, in: I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, R. Garnett (Eds.), Advances in Neural Information Processing Systems, volume 30, Curran Associates, Inc., 2017. URL: https://proceedings.neurips.cc/paper_files/paper/2017/file/ 7c82fab8c8f89124e2ce92984e04fb40-Paper.pdf. [33] P. H. Cabot, R. Navigli, REBEL: relation extraction by end-to-end language generation, in: M. Moens, X. Huang, L. Specia, S. W. Yih (Eds.), Findings of the Association for Computational Linguistics: EMNLP 2021, Virtual Event / Punta Cana, Dominican Republic, 16-20 November, 2021, Association for Computational Linguistics, 2021, pp. 2370– 2381. URL: https://doi.org/10.18653/v1/2021.findings-emnlp.204. doi: 10.18653/v1/2021. findings-emnlp.204. [34] Y. Lu, Q. Liu, D. Dai, X. Xiao, H. Lin, X. Han, L. Sun, H. Wu, Unified structure generation for universal information extraction, in: S. Muresan, P. Nakov, A. Villavicencio (Eds.), Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2022, Dublin, Ireland, May 22-27, 2022, Association for Computational Linguistics, 2022, pp. 5755–5772. URL: https://doi.org/10.18653/v1/2022. acl-long.395. doi:10.18653/v1/2022.acl-long.395. [35] S. Wadhwa, S. Amir, B. C. Wallace, Revisiting relation extraction in the era of large language models, in: A. Rogers, J. L. Boyd-Graber, N. Okazaki (Eds.), Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2023, Toronto, Canada, July 9-14, 2023, Association for Computational Linguistics, 2023, pp. 15566–15589. URL: https://doi.org/10.18653/v1/2023.acl-long.868. doi:10.18653/v1/2023.acl-long.868. [36] A. Galassi, M. Lippi, P. Torroni, Attention in natural language processing, IEEE Trans.
Neural Networks Learn. Syst. 32 (2021) 4291–4308. URL: https://doi.org/10.1109/TNNLS. 2020.3019893. doi:10.1109/TNNLS.2020.3019893.