Neural Networks Over Neo4j

1. Introduction

Twitter mining analytics provide a significant oppor(E. Kafeza) (E. Kafeza) CIKM’22: 31st ACM International Conference on Information and CEUR htp:/ceur-ws.org ISN1613-073

CEUR

Workshop Proceedings (CEUR-WS.org) https://www.imdb.com/title/tt0133093 sion making processes and marketing performance. This is achieved as businesses are allowed through Twitter mining to gain invaluable insights on the dynamics and collective behavior of their customer base or any other online target group for that matter. In turn this yields more accurate predictions of key factors such as future demand, reactions to new products, or brand loyalty to name only a few. Twitter analytics tailored for this task include community structure discovery algorithms, hashtag flow analysis and information difusion strategies, digital influence computation, and link prediction tools.

Such insight is obtained frequently from the computationally challenging task of processing a diverse set of follow relationships, hashtags, or tweets. Such attributes are either of structural nature in the sense that they are about the social graph itself or functional as they pertain to the activity of the various entities, mostly the Twitter accounts, which use said graph.

Among the functional features the afective ones have recently garnered considerable research attention since emotions are the primary motivations behind human larity of tweets and hashtags are considered as major indicators of how a Twitter account would react to various events and rely heavily on emotion models such as those proposed by Plutchik or Ekman. However, personality taxonomies such as the Myers-Brigs taxonomy indicator (MBTI) go beyond individual emotional responses and provide a more general framework for systematically evaluating sequences of account reactions as they take them. For instance, personalities with an extrovert pretunity to various organizations to improve both deci- actions. To this end, attributes such as the emotional po© 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License into consideration the higher cognitive functions driving Attribution 4.0 International (CC BY 4.0). disposition will be typically more vociferous compared ture extending as a result the digital awareness of the to introvert ones for the same event. 5. Sets are denoted by capital letters. Bold capital letters denote matrices, small bold vectors, and small scalars.

Acronyms are explained the first time they are encountered in the text. In function definitions parameters are separated of the arguments by a semicolon. Finally, in

2. Related

Work

IA design relies on a number of approaches [1] as they have to function on their own in large digital infrastrucorganizations deployed them. IAs often have to take decisions [2, 3] which in turn rely on operational criteria based on aspects like anthropomorphism [4], maintaining trust with human users [5], cognitive functionality [6], connecting IAs with sensors [7], action explainability [8], and even the possible role of voice [9]. IAs can be employed in many capacities like protecting critical industrial cyber-physical infrastructure [10], communicating with humans through dynamic oral conversations [11], facilitating social interactions [12], recommending charging points for electric vehicles [13], modeling financial markets [14], and even shaping fashion trends [15].

In order for IAs to adapt to complex and nonstationary environments, ML capabilities have been recently added to them [16]. ML techniques cover a broad spectrum of options such as variational encoders [17], reinforcement learning [18], deep learning in various forms [19], and cooperative learning [20]. Possible extensions for use with IAs are tensor stack networks (TSNs) [21], GNNs [22], adversarial neural networks (ANNs) [23], and self organizing maps (SOMs) [24].

Graph mining extracts nontrivial knowledge from graphs with undirected ones based on density criteria [26]. Community discovery for large graphs has taken many forms due not only to instance size [27] but also because many and equally valid graph community definitions exist [28, 29]. For instance, communities may well be build on trust [30], spatiotemporal patterns [31], social behavior [32], multiple connectivity criteria [33], noiseless patterns [34], spatial behavior akin to that of geolocation services [35], privacy preserving constraints [36], and simultaneous structural and afective criteria [37]. Applications include trajectory planning for autonomous race vehicles [38], political [39] and commercial [40] digital campaign designs, biomedical document recommendation based on a keyword-term-document tensor model [41], opinion mining [42], and assessing

Personality indicators [45] go beyond emotion models like the emotion wheel [46] or big five

[47] since they provide a framework for predicting reactions to a wider array of stimuli [48], whereas the various emotion models only describe a single reaction. MBTI is used among others for designing digital campaigns on social media [49] perhaps in conjunction with other major attributes of the underlying domain [50]. Also MBTI taxonomy has been used to assess leadership traits [51]. Examples include exploring political Twitter [52], predicting brand loyalty [53], and analyzing the dynamics in social media communities [54]. Tensor distance metrics [55] can be used to cluster personalities in the MBTI space, especially when personalization is intended [56]. jump based on a strategy exploiting local structural in- linked data [25]. For example, approximating directed reviewed. IA design is described in detail in section 3. the afective resilience of Twitter graphs [ 43, 44]. 3. Intelligent Agent Design 3.1. Graph representation label in carries a diferent semantic value and denotes relationships of varying strength. In the proposed approach this is translated to diferent edges having priority depending on their label, but this does not exclude the association of specific edge weights with diferent labels.

△ = ( , ,

) Multilayer graphs in this work will be the algorithmic cornerstone for representing and analyzing the interaction between Twitter accounts and the IA as the latter will be circulating along the edges of such a graph. Multilayer graphs extend ordinary ones by including a set of distinct edge labels along with the requirement that an edge has exactly one and, thus, allowing as many as multiple edges between any two vertices as long as their labels are distinct. As such, concepts such as vertex degrees and paths have to be redefined to take into consideration this extension. As each label carries its own structural, functional, and semantic meaning, many underlying domains can be better modeled. Definition 1 formally introduces the class of labeled multilayer graphs.

As mentioned earlier, MBTI is one of the most popular personality models stemming in large part from the theory of Jung and it aims at assessing the combination of individual cognitive functions. Specifically, in MBTI there are four independent axes, each corresponding to a function, with their two endpoints being personality attributes. This yields a total of sixteen basic personality types known by the initials of their corresponding attributes as shown in figure 1. Therein the respective frequencies are also shown. Observe that in this case as well holds the fundamental result of probability theory Definition 1 (Labeled multilayer graphs). A labeled stating that for every discrete distribution with data multilayer graph expresses simultaneous connections points there is at least one such point whose probability between its vertices allowing the formulation of higher is strictly more than 1/ . In the context of MBTI this order patterns and it is defined as the ordered triplet of ( 1 ): implies there is at least one personality type which is more common than the others.

3.2. MBTI personality system

( 1 )

The four axes representing basic cognitive functions or fundamental personality traits which determine the MBTI taxonomy are the following: • Extroversion vs Introversion: This axis determines how social an individual is. In turn this determines the nature and amount of sensory input an individual can cognitively process. • Sensing vs iNtuition: This direction indicates the role tangible data play in a person’s cognitive pro

The components of a multilayer and the role they play in representing Twitter accounts and the associated interactions and activities are the following: • The set of vertices comprises of the entities the relationships are built on. In the context of the proposed methodology consists of anonymized

Twitter accounts. • The set ⊆ × × is the set of labeled edges.

Therefore, not only does each edge have orientation but also a label. In this work they denote

Twitter interaction. • The label set depends directly on the semantics of the underlying domain. In this particular case, contains the four elements of equation ( 2 ).

=△ { , , , }

( 2 )

Definition 1 takes into consideration structural and functional graph characteristics. The former rely on combinatorial properties, whereas the latter depend directly on the Twitter activity. For the purposes of the analysis done here edges have directions, capturing the one-way nature of Twitter interaction.

Each of the Twitter graph layers is formed by a specific label and their endpoints. A given layer need not be symmetric and, in fact, such a layer is a special case. Each

Each layer is formed by the edges of a single label, Figure 1: MBTI personalities. therefore allowing in total | | layers. Each of them represents diferent account interaction patterns.

ISTJ 12% ISTP 4% ESTP 5% ESTJ 12%

ISFJ 8% ISFP 3% ESFP 5% ESFJ 8%

INFJ 4% INFP 4% ENFP 8% ENFJ 5%

INTJ 7% INTP 4% ENTP 5% ENTJ 6% cess, namely whether they rely more on concrete input or abstract terms. • Feeling vs Thinking: This variable denotes whether an individual tends to reason in order to understand a situation or relies on accumulated experience in the form of hunches. • Perceiving vs Judging: This factor finally shows whether a person tends to place themselves inside a given situation during decision making process or they reason from a detached standpoint.

The GNN performing the vertex classification works as follows. It consists of 0 layers where each one acts like a local spatial filter applied in parallel on the various graph segments similarly to a convolutional neural network (CNN). This is achieved through successive applications of the same nonlinear mapping (⋅), frequently reported as the activation function, on a linear transform of the graph adjacency matrix. In this equation P is the graph Laplacian matrix of ( 4 ), H is the output of the previous layer, and W is a trainable weight matrix.

The system described above captures a major part of M+1 = (PH W ) ( 3 ) human decision making process as it describes how and which information is collected and furthermore how it is In ( 3 ) the graph Laplacian matrix P is defined as in ( 4 ), processed. As stated earlier the personalities of the MBTI where A is the graph adjacency matrix and D is the graph taxonomy are shown in figure 1 in such a way that ech neighborhood size matrix, namely the diagonal matrix personality difers at exactly one trait from its neighbor- containing the total number of labeled edges whose ending ones. This property in fact extends to personalities point is the respective vertex. Since Twitter graphs are which are adjacent if the taxonomy map would wrap. directed, then there are two adjacency matrices, one for Thus, personalities are encoded in a scheme similar to the inbound and one for the outbound edges. Moreover, that of Gray coding allowing the easy grouping by factors since in this work there can be multiple edges between similar to Karnaugh maps. the same vertex pair, both A and D contain the count of

In order to estimate the MBTI personality of each Twit- edges whose head or tail is the respective vertex. This ter account in the dataset a GNN was used which oper- implies that also the diagonal of A should be bolstered ated on the entire graph. Since each of the four variables with the maximum number of parallel edges, since each of the taxonomy are independent, their estimation was vertex is strongly connected with itself. Instead of usrecast as four separate instances of vertex classification. ing an one-hot encoding for each of personality type, a The ground truth state vector contains the following vector with four entries was used as it is a more natural attributes, which can be linked to how the basic person- representation for the MBTI taxonomy. alities of the MBTI taxonomy manifest themselves.

• Average number of connections. • Average number of characters. • Average number of nonwhite characters. • Fraction of alphabetical characters. • Fraction of digits. • Fraction of uppercase characters. • Fraction of white spaces. • Fraction of special characters. • Average number of words. • Fraction of unique works. • Number of long words. • Average word length. • Number of unique stopwords. • Fraction of stopwords. • Number of sentences. • Number of long sentences (at least 10 words). • Average number of characters per sentence. • Average number of words per sentence. • Percentage of positive words. • Percentage of negative words. • Percentage of neutral words.

As a result of the above, there are two CNNs, one operating on inbound and another on outbound neighborhoods. The nonlinear activation function (⋅)shown in ( 5 ) is applied elementwise to the matrix argument.

( )=△ ln ( 0 + 1)

The synaptic weights of matrix W are also individually updated using the delta rule of equation ( 6 ). Specifically, each element of the weight matrix in the -th layer receives a correction term of the form: ΔW [, ] = 0( 0 0M+1 [, ] 1 + 0M+1 [, ] )M+1 [, ]

( 6 ) The learning parameter 0 decays with a cosine rate whose frequency 0 depends on the training size 0.

When both CNN run and the personality types are obtained, then in case there is a diference between the two CNNs for the value a specific variable, it is decided by the majority of the values of the respective variables of the neighboring vertices ignoring direction. When this is not possible, then second order neighborhoods are also included, again ignoring direction.

( 4 ) ( 5 )

3.3. Jump strategy

The IA proposed here moves along the edges of the graph following a heuristic mechanism for community structure discovery which eventually approximates under a set of mild assumptions a homogeneous Markov chain steady state distribution. IA starts from an arbitrary vertex and visits other vertices with probability proportional to their inbound degree locally normalized. Progressively this constructs a long sequence ⟨ ⟩ containing the vertices visited by the IA, which can then be mined by any agglomerative clustering algorithm in order for the final community list to be derived. Longer sequences lead to more reliable discovery as they contain a higher variety of subsequences. Thus there are more indications whether certain vertices belong together. In particular if triangles local are successfully identified, then the more likely a community is to be properly discovered. Additional patterns assisting in community discovery are:

nected with at least one bridge. • Articulations: They are bridge endpoints. As such, they are critical for connectivity and belong to diferent communities. • Subsequences: Frequent subsequences are indicators that certain vertices should be grouped together, especially for shorter ones. munities. As such, they should be grouped with the vertices they appear with.

In the limit the IA random walk approximates the steady state distribution of a homogeneous Markov chain as the vertex selection mechanism is solely determined by the local connectivity patterns between the current and the outbound neighboring vertices. This also eliminates the efect of the starting vertex.

In order to determine the next vertex to be visited the IA makes a probabilistic decision based on a mechanism reminiscent of preferential attachment. Recall that according to the latter the probability of moving from vertex to an outbound neighbor is shown in equation ( 7 ). Thus, each candidate vertex is selected with a probability proportional to its locally normalized inbound degree. The normalization constant is the sum of the in-degrees of every outbound neighbor of .

|Γ ( )| ∑ ∈Γ ( ) |Γ ( )| =

1 1 + ( )

( 7 )

In ( 7 ) the positive quantity (⋅) which is a function of the vertex and of the union of the inbound neighborhoods of the outbound neighborhood of the current • Bridges: Losing a bridge always implies connec

In order to derive a probability distribution from the tivity loss. Thus, two communities may be con- scores obtained by ( 7 ), the softmax map is used in this • Hubs and authorities: Both are central in com- ( 9 ), any decision rule determining the next vertex IA vertex is obtained if both the numerator and the denominator of equation ( 7 ) are divided by |Γ ( )|.

( )= △

∑ ∈Γ ( )⧵ |Γ ( )| |Γ ( )|

If the middle form of equation ( 7 ) is used, then |Γ ( )| can be eficiently approximated under mild conditions by large set cardinality estimators [57]. Alternatively, the right hand side form of ( 7 ) is an inspiration for approximating ( ) in ( 8 ) by ̂ ( ), when now the ratio of the estimators is used or, less frequently, an estimation of their ratio since ratio statistics are in general dificult to be derived. In any case, this needs to be done only once for static graphs. In that case the rightmost form of equation ( 7 ) can be approximated as in equation ( 9 ):

1 1 + ̂ ( ) =

1 1 + ln ̂ ( ) = 1 − (ln ̂ ( ))

( 9 ) work. Specifically, the transition probability from to an outbound neighbor can be computed as: prob { → } =

exp (1 − (ln ̂ ( ))) ∑ exp (1 − (ln ̂ ( )))

, ∈ Γ ( )

Alternatively, given the probabilistic approximation moves to which relies on the ratio of the transition probabilities from to two distinct outbound neighbors and can use the equivalent computation of equation ( 11 ): prob { → } prob { → } = 1 − (ln ̂ ( )) 1 − (ln ̂ ( ))

In ( 9 ) (⋅)is the logistic function defined in equation ( 12 ). It is always positive and also it is the derivative of the softplus function commonly used as the nonlinear activation function in neural networks [58].

= = ( )= △

1 1 + − + 1

An important property of the logistic function which has been used in ( 9 ) is shown in ( 13 ). This property suggests a balance or a conservation law between the logistic function and its symmetric with respect to its argument and the horizontal axis = 1/2 .

(− ) =

1 1 +

− 1 + − = 1 − ( )

( 13 )

Moreover, notice that (⋅) is a rescaled and shifted version of the hyperbolic tangent (⋅)of ( 14 ). The latter is the Bayesian estimator of a bipolar ±1 bit in the presence of additive Gaussian white noise [59]. Thus the logistic function is the Bayesian estimator for a 0/1 bit.

( )=△ tanh ( )= △ − − + −

The absolute derivative value of the logistic function is bound as shown in equation ( 15 ). This means that any approximation errors of ̂ are not magnified by (⋅).

( ) | | = |−

− − (1 + − )2 | = | ( ) (− )| ≤ 1

( 15 )

Extending ( 7 ) in order to include multiple labels gives ( 18 ). Therein the transition probability from to takes into account the parallel edges wherever they exist with additional sums raging over . Alternatively, this can be replaced by the cardinality of the set of edges of the form (, ;

)for the various possible labels ∈ . Additionally, let locations of the MBTI personalities of and in figure 1 be ( , ) and ( , ). Then the factor , depends on these locations is computed as follows:

( 16 ) which , =△ exp (− ( − )2 + ( − )2

) 2

A last but optional modification to ( 7 ) is the addition of a forgetting factor ,

which penalizes outbound neighbors which are selected too frequently in favor of others. tions is that the IA should be able to escape very dense segments of a particular community in order to explore alternative paths inside the community or even move to other communities. The particular form of the forgetting factor used here is shown in equation ( 17 ). Therein , is the number of times has been selected as a destination for the IA from the last 0 times.

△ , = 1 − , 1 + 0

With these observations, the original decision rule of equation ( 7 ) is now modified to take into consideration the fact that the IA moves along a multilayer graph where vertices have their own personality according to the MBTI taxonomy giving equation ( 18 ). Moreover, the forgetting factor is also present, but its use is optional.

The rationale for taking into consideration past selec- can be cached such that (⋅) can be eficiently approxithe selection of parameter 0 there will be suficient visits to obtain important information regarding the role of each vertex. Because of the decision rule ( 18 ), vertices which are the endpoints of important edges will be visited more frequently, whereas vertices peripheral in large communities, with low degree, or dificult to reach will be progressively neglected up to a point.

0 = | | log 0 | |

4. Results

4.1. Implementation

In figure 2 the components of the proposed system are shown. Therein can be seen that the IA moves on a multilayer graph stored in a standalone Neo4j instance along with the activity on it such as tweets and mentions. Another major element is the GNN which has been executed as a preprocessing step and has given to each vertex its MBTI personality. The third component is the IA itself, which is lightweght as it only has to implement the decision rule of ( 18 ) based on local input.

The operational parameters of the GNN and the IA are shown in table 2 and they pertain to various equations presented earlier in the text.

The dataset used in this work has been collected by the Twitter crawler used among others in [26]. The data therein pertains to three diferent graphs constructed from topic sampling using a main hashtag. Specifically, the three Twitter graphs are the following: Neo4j +follow ->retweet #hashtags jump decisions

MBTI estimations language. At the sampling time the next Julia de- figure over, earlier that year a major language update was released. Therefore, there was high interest at the time for the particular topic with mostly positive or neutral feelings. • #Windows11: At about the same time a major news update about the upcoming Windows 11 was made to the public. This generated mostly positive sentiment, but also some negative ones concerning the removal of some of the expected ware support still lingering among users. • #BlackList: Just before the beginning of the final season of this the major hit it was announced that the main protagonist would not join the cast. Furthermore this was aggravated by leaked plots where her character is removed in a way deemed anticlimactic by fans. These stirred considerable controversy among them.

To form the CNN ground truth from the collected Twitter dataset two diferent ways were used: • By picking the MBTI type an account has posted on their Twitter bio. This is the preferred way and it was used whenever it was possible. • By locating a reference to an MBTI personality up to two words apart from the words I, am, feel, myself, me, and being.

Accounts with self-reported MBTI type were not included in the vertex classification by the CNN.

4.2. Evaluation

The Kullback-Leibler divergence between the distribution of the MBTI types returned by CNN for each graph compared to the global reference distribution shown in 1 is shown in table 4. Recall that the divergence between a distribution and a reference one when they are both discrete is computed by equation (20). ⟨|| ⟩ = ∑ log ( △

) (20) The low divergence between the three empirical distributions and the reference one mean that the three graphs are representative in terms of MBTI personalities.

Hashtag coherency is a functional way to assess the this approach is that a successful partitioning will result in communities of accounts with similar interests as expressed by hashtags. Specifically, if the Tanimoto of merit. The results can be seen in table 5. coeficient 0 between the hashtag sets of two accounts is used as a distance metric, then the minimum and the average intercluster distances can be used figures

Table 5 should be read as follows. Three cases were tested. The first was the decision rule of ( 18 ) without the MBTI similarity factor , and the forgetting factor , . The second was the same rule with only , and the third was with both , and , . For all three cases and were collected and each was normalized with respect to its respective maximum. This allows the percentage of change between jump strategies to be shown. In light of this, enabling both factors result to better communities, while excluding them both yields the worst ones.

Finally, the role of forgetting factor , it was positive as it can be seen from the above metrics. This can be attributed to the fact that, although at first iy may look counter-intuitive, it has a linear and not an exponential decay, so it forces the IA to choose less likely outbound neighbors but not too often.

5. Future

Work

This conference paper focuses on the development of an intelligent agent (IA) which performs random walks on Twitter multilayer graphs with the purpose of estimating edge frequencies, which in turn heavily rely on a local decision rule for selecting the destination vertex. Moreover, destination vertices with similar MBTI profile types are given priority, whereas vertices frequently selected may be optionally penalized in order to allow the IA to escape from especially dense segments of the Twitter graph. Experimental results indicate that both factors increase the partitioning quality in terms of increased minimum and average intercluster distance.

Concerning future research directions, the most immediate one is to apply the proposed approach on more and larger benchmark Twitter graphs. Moreover, IAs may work in parallel in cooperative or adversarial modes.

Acknowledgments

This conference paper is part of Project 451, a long term research initiative with a primary objective of developing novel, scalable, numerically stable, and interpretable higher order analytics.

Moreover, this work is part of Project Vega.

References

[1] R. Calegari , G. Ciatto , V. Mascardi , A. Omicini , Logic-based technologies for multi-agent systems: A systematic literature review , Autonomous Agents and Multi-Agent Systems 35 ( 2021 ) 1 - 67 .

[2] D. Ha , Y. Tang , Collective intelligence for deep learning: A survey of recent developments , Collective Intelligence 1 ( 2022 ).

[3] I. M. Enholm , E. Papagiannidis, P. Mikalef , J. Krogstie , Artificial intelligence and business value: A literature review , Information Systems Frontiers ( 2021 ) 1 - 26 .

[4] S. Moussawi , M. Koufaris , R. Benbunan-Fich , How perceptions of intelligence and anthropomorphism afect adoption of personal intelligent agents , Electronic Markets 31 ( 2021 ) 343 - 364 .

[5] T. Kim , H. Song , How should intelligent agents apologize to restore trust? Interaction efects between anthropomorphism and apology attribution M. Dunion , E. Fosong , S. Garcin , S. Guo , B. Gyevnar , on trust repair, Telematics and Informatics 61 T. McInroe , et al., Deep reinforcement learning for (2021). multi-agent interaction , AI Communications ( 2022 )

[6] Z. Nagoev , I. Pshenokova , O. Nagoeva , Z. Sundukov , 1 - 12 . Learning algorithm for an intelligent decision mak- [20] I.-J. Liu , U. Jain , R. A. Yeh , A. Schwing , Cooperaing system based on multi-agent neurocognitive ar- tive exploration for multi-agent deep reinforcement chitectures , Cognitive Systems Research 66 ( 2021 ) learning , in: ICML, PMLR, 2021 , pp. 6826 - 6836 . 82 - 88 . [21] G. Drakopoulos , E. Kafeza, P. Mylonas , L. Iliadis,

[7] A. Gupta , S. Savarese , S. Ganguli , L. Fei-Fei , Embod- Transform-based graph topology similarity metied intelligence via learning and evolution, Nature rics , NCAA 33 ( 2021 ) 16363 - 16375 . doi:1 0 . 1 0 0 7 / communications 12 ( 2021 ) 1 - 12 . s 0 0 5 2 1 - 0 2 1 - 0 6 2 3 5 - 9 .

[8] G. Vilone , L. Longo, Notions of explainability and [22] X. Zhou , W. Liang , W. Li , K. Yan , S. Shimizu , I. Kevin, evaluation approaches for explainable artificial in- K. Wang, Hierarchical adversarial attacks against telligence , Information Fusion 76 ( 2021 ) 89 - 106 . graph neural network based IoT network intrusion

[9] K. Seaborn , N. P. Miyake , P. Pennefather , M. Otake- detection system , IEEE Internet of Things Journal Matsuura , Voice in human-agent interaction: A ( 2021 ). survey, CSUR 54 ( 2021 ) 1 - 43 . [23] F. E. Fernandes Jr , G. G. Yen , Pruning of generative

[10] Z. Lv , D. Chen , R. Lou , A. Alazab , Artificial intelli- adversarial neural networks for medical imaging gence for securing industrial-based cyber-physical diagnostics with evolution strategy, Information systems , Future generation computer systems 117 Sciences 558 ( 2021 ) 91 - 102 . ( 2021 ) 291 - 298 . [24] G. Drakopoulos , I. Giannoukou, S. Sioutas , P. My-

[11] E. C. Ling , I. Tussyadiah , A. Tuomi , J. Stienmetz, lonas, Self organizing maps for cultural conA . Ioannou, Factors influencing users' adoption and tent delivery , NCAA 31 ( 2022 ). doi:1 0 . 1 0 0 7 / use of conversational agents: A systematic review , s 0 0 5 2 1 - 0 2 2 - 0 7 3 7 6 - 1 . Psychology & Marketing 38 ( 2021 ) 1031 - 1051 . [25] H. Motoda , K. Saito , Critical link identification

[12] S. K. Lee , P. Kavya , S. C. Lasser , Social interactions based on bridge detection for network with uncerand relationships with an intelligent virtual agent, tain connectivity , in: ISMIS , Springer, 2018 . International Journal of Human-Computer Studies [26] G. Drakopoulos , E. Kafeza, P. Mylonas , S. Sioutas, 150 ( 2021 ). Approximate high dimensional graph mining with

[13] W. Zhang , H. Liu, F. Wang , T. Xu , H. Xin , D. Dou, matrix polar factorization: A Twitter application , H. Xiong, Intelligent electric vehicle charging rec- in: IEEE Big Data , IEEE, 2021 , pp. 4441 - 4449 . doi: 1 0 . ommendation based on multi-agent reinforcement 1 1 0 9 / B i g D a t a 5 2 5 8 9 . 2 0 2 1 . 9 6 7 1 9 2 6 . learning , in: The Web Conference, ACM, 2021 , pp. [27] G. Mei , X. Wu , Y. Wang , M. Hu , J.-A. Lu , G. Chen, 1856 - 1867 . Compressive-sensing-based structure identification

[14] J. Lussange , I. Lazarevich, S. Bourgeois-Gironde, for multilayer networks , IEEE transactions on cyS. Palminteri, B. Gutkin, Modelling stock markets bernetics 48 ( 2017 ) 754 - 764 . by multi-agent reinforcement learning , Computa- [28] X. Su , S. Xue , F. Liu , J. Wu , J. Yang , C. Zhou , W. Hu, tional Economics 57 ( 2021 ) 113 - 147 . C. Paris, S. Nepal, D. Jin , et al., A comprehensive

[15] O. Allal-Chérif , V. Simón-Moya , A. C. C. Ballester, survey on community detection with deep learnIntelligent purchasing: How artificial intelligence ing, IEEE Transactions on Neural Networks and can redefine the purchasing function , Journal of Learning Systems ( 2022 ). Business Research 124 ( 2021 ) 69 - 76 . [29] G. Rossetti , R. Cazabet , Community discovery in

[16] S. Carta , A. Ferreira , A. S. Podda , D. R. Recupero , dynamic networks: A survey , CSUR 51 ( 2018 ) 1 - 37 . A. Sanna , Multi-DQN : An ensemble of deep Q- [30] G. Drakopoulos , E. Kafeza, P. Mylonas , learning agents for stock market forecasting, Expert H . Al Katheeri, Building trusted startup teams from systems with applications 164 ( 2021 ). LinkedIn attributes: A higher order probabilistic

[17] B. Wu , S. Nair , R. Martin-Martin , L. Fei-Fei , C. Finn, analysis , in: ICTAI, IEEE, 2020 , pp. 867 - 874 . Greedy hierarchical variational autoencoders for doi:1 0 . 1 1 0 9 / I C T A I 5 0 0 4 0 . 2 0 2 0 . 0 0 1 3 6 . large-scale video prediction , in: CVPR, IEEE/CVF, [31] S. van den Beukel, S. H. Goos , J. Treur , An adaptive 2021 , pp. 2318 - 2328 . temporal-causal network model for social networks

[18] M. L. Olson , R. Khanna , L. Neal , F. Li , W.-K. Wong, based on the homophily and more-becomes-more Counterfactual state explanations for reinforce- principle , Neurocomputing 338 ( 2019 ) 361 - 371 . ment learning agents via generative deep learning , [32] P. Bindu , P. S. Thilagam , D. Ahuja , Discovering Artificial Intelligence 295 ( 2021 ). suspicious behavior in multilayer social networks,

[19] I. H. Ahmed , C. Brewitt , I. Carlucho , F. Christianos , Computers in Human Behavior 73 ( 2017 ) 568 - 582 .

[33] M. Abdolali , N. Gillis , M. Rahmati , Scalable and [47] Q. Agbaria , A. A. Mokh , Coping with stress during robust sparse subspace clustering using randomized the coronavirus outbreak: The contribution of big clustering and multilayer graphs, Signal Processing ifve personality traits and social support , Interna163 ( 2019 ) 166 - 180 . tional Journal of Mental Health and Addiction 20

[34] G. Drakopoulos , E. Kafeza, P. Mylonas , S. Sioutas , ( 2022 ) 1854 - 1872 . Process mining analytics for Industry 4 .0 with [48] J. Juárez , C. Santos , C. A. Brizuela , A comprehensive graph signal processing, in: WEBIST, SCITEPRESS, review and a taxonomy proposal of team formation 2021 , pp. 553 - 560 . doi:1 0 . 5 2 2 0 / 0 0 1 0 7 1 8 3 0 0 0 0 3 0 5 8 . problems, CSUR 54 ( 2021 ) 1 - 33 .

[35] Y. Xie , M. Gong , S. Wang , B. Yu , Community discov- [49] N. Cerkez , B. Vrdoljak , S. Skansi , A method for ery in networks with deep sparse filtering , Pattern MBTI classification based on impact of class comRecognition 81 ( 2018 ) 50 - 59 . ponents, IEEE Access 9 ( 2021 ) 146550 - 146567 .

[36] X. Zheng , Z. Cai , G. Luo , L. Tian , X. Bai , Privacy- [50] K. A. Nisha , U. Kulsum , S. Rahman , M. Hossain, preserved community discovery in online social P . Chakraborty, T. Choudhury , et al., A comparnetworks, Future Generation Computer Systems ative analysis of machine learning approaches in 93 ( 2019 ) 1002 - 1009 . personality prediction using MBTI , in: Computa-

[37] D. Drakopoulos , K. C. Giotopoulos , I. Giannoukou, tional Intelligence in Pattern Recognition, Springer, S. Sioutas, Unsupervised discovery of semantically 2022 , pp. 13 - 23 . aware communities with tensor Kruskal decompo - [51] P. E. Roush , L. Atwater , Using MBTI to understand sition: A case study in Twitter, in: SMAP, IEEE, transformational leadership and self-perception ac2020 . doi:1 0 . 1 1 0 9 / S M A P 4 9 5 2 8 . 2 0 2 0 . 9 2 4 8 4 6 9 . curacy, Military Psychology 4 ( 1992 ) 17 - 34 .

[38] T. Stahl , A. Wischnewski , J. Betz , M. Lienkamp , [52] A. Khatua , A. Khatua , E. Cambria, Predicting politiMultilayer graph-based trajectory planning for race cal sentiments of voters from twitter in multi-party vehicles in dynamic scenarios , in: ITSC, IEEE, 2019 , contexts, Applied Soft Computing 97 ( 2020 ). pp. 3149 - 3154 . [53] L. Lalicic , A. Huertas , A. Moreno , M. Jabreel , Emo-

[39] V. K. Brändle , C. Galpin , H. - J. Trenz , Brexit as 'pol- tional brand communication on Facebook and Twititics of division': Social media campaigning after ter: Are DMOs successful? , Journal of Destination the referendum, Social Movement Studies 21 ( 2022 ) Marketing & Management 16 ( 2020 ). 234 - 253 . [54] E. J. Choong , K. D. Varathan , Predicting judging-

[40] L. Dolega , F. Rowe , E. Branagan, Going digital? The perceiving of Myers-Briggs Type Indicator (MBTI) impact of social media marketing on retail website in online social forum , PeerJ 9 ( 2021 ). trafic, orders and sales , Journal of Retailing and [55] G. Drakopoulos , I. Giannoukou , P. Mylonas , Consumer Services 60 ( 2021 ). S. Sioutas, On tensor distances for self organiz-

[41] G. Drakopoulos , A. Kanavos , I. Karydis, S. Sioutas, ing maps: Clustering cognitive tasks , in: DEXA, A. G. Vrahatis , Tensor-based semantically- aware volume 12392 , Springer, 2020 , pp. 195 - 210 . doi: 1 0 . topic clustering of biomedical documents , Compu- 1 0 0 7 / 9 7 8 - 3 - 0 3 0 - 5 9 0 5 1 - 2 _ 1 3 . tation 5 ( 2017 ). doi:1 0 . 3 3 9 0 / c o m p u t a t i o n 5 0 3 0 0 3 4 . [56] D. Fernau , S. Hillmann , N. Feldhus , T. Polzehl,

[42] J. Serrano-Guerrero , F. P. Romero , J. A. Olivas , Towards automated dialog personalization using Fuzzy logic applied to opinion mining: A review, MBTI personality indicators , Proc. Interspeech 2022 KBS 222 ( 2021 ). ( 2022 ) 1968 - 1972 .

[43] G. Drakopoulos , E. Kafeza, P. Mylonas , S. Sioutas, [57] G. Drakopoulos , S. Kontopoulos , C. Makris, EvenA graph neural network for fuzzy Twitter graphs, tually consistent cardinality estimation with appliin: G . Cong, M. Ramanath (Eds.), CIKM companion cations in biodata mining , in: SAC, ACM, 2016 . volume, volume 3052 , CEUR-WS.org, 2021 . doi:1 0 . 1 1 4 5 / 2 8 5 1 6 1 3 . 2 8 5 1 8 8 7 .

[44] G. Drakopoulos , I. Giannoukou , P. Mylonas, [58] Y. Chen , Y. Mai , J. Xiao , L. Zhang , Improving the S. Sioutas , A graph neural network for assessing antinoise ability of DNNs via a bio-inspired noise the afective coherence of Twitter graphs, in: IEEE adaptive activation function rand softplus, Neural Big Data , IEEE, 2020 , pp. 3618 - 3627 . doi:1 0 . 1 1 0 9 / computation 31 ( 2019 ) 1215 - 1233 . B i g D a t a 5 0 0 2 2 . 2 0 2 0 . 9 3 7 8 4 9 2 . [59] S. Cheng, H. Zhan, H. Yao , H. Fan , Y. Liu, Large-

[45] S. Štajner , S. Yenikent , Why is MBTI personality scale many-objective particle swarm optimizer with detection from texts a dificult task?, in: European fast convergence based on alpha-stable mutation Chapter of the Association for Computational Lin- and logistic function , Applied Soft Computing guistics: Main Volume, 2021 , pp. 3580 - 3589 . ( 2020 ) 106947 .

[46] X. Zeng , Q. Chen , X. Fu , J. Zuo , Emotion wheel attention-based emotion distribution learning , IEEE Access 9 ( 2021 ) 153360 - 153370 .