3 of 79

Mining Text Data

Over 80% of the data is in an unstructured format, e.g., news, image, audio.
Goal: deriving high-quality information, such as structures, patterns, and summaries, from unstructured text data.

Process: mining structures from text ⇒ deriving patterns within the structured data ⇒ evaluating and interpreting output

Main challenge:

How to represent text primitives (e.g., words, sentences, and documents) and how to learn effective representation from unstructured text?

4 of 79

Mining Text Data: Major Tasks

Information extraction (IE): automatically extracting structured information from unstructured and semi-structured documents.
Taxonomy construction: constructing different multi-faceted taxonomies based on different corpora or applications.
Text clustering: grouping unlabeled texts with similar contextual patterns.
Text classification: analyzing open-ended texts (e.g., webpages) and

assigning them a set of predefined topics or categories

Text summarization: creating a subset of text that represents the most important information from a single or multiple pieces of text.

5 of 79

Mining Text Data: Techniques

Symbol-based representation: using one-hot vectors for words and bag-of-words model for documents.

Issues: high-dimensionality and data sparsity

Distributed representation: low-dimensional real-valued dense vectors with large amount of unlabeled data using deep neural network.

Word embedding: words similar in meaning are expected to be closer in the embedded space.
Pre-trained language model: using large-scale data and more computing resources to generate contextualized embeddings.

6 of 79

Mining Text Data: Techniques

Word2vec: words that share common contexts are embedded in close proximity, e.g., “Sweden" can be embedded close to Norway, Finland.

CBOW: predicts the current word from surrounding context words.
Skip-gram: uses the current word to predict the surrounding words.

GloVe: combining global matrix factorization

and local context window methods.

JoSE: learning word embeddings in hyperbolic

space to model hierarchical structures.

Pre-trained language model: ELMo and BERT

7 of 79

Mining Rich Data Types

Mining Text Data
Mining Spatial-Temporal Data
Mining Graph and Networks

8 of 79

Mining Spatial-Temporal Data: Properties and Types

Key Properties in Spatial and Temporal Data

Auto-correlation: dependencies and correlations in proximate locations and time, e.g., similar traffic status for adjacent intersections.
Heterogeneity: heterogeneous and non-stationary in distribution.

Types of Spatial and Temporal Data

Data type	Definition	Operation	Examples
Event	at a spatial location and a time point	intersections and similarity/difference	intersection of forest fire and a rainfall returns the overlapping area and time
Trajectory	a path of an object over space and time	finding relationships between two trajectories	transportation (e.g., tracing vehicle), epidemiology, ecology
Point reference data	collected using discrete reference points	reconstructing fields and modeling non-stationary random process	using mobile sensors to collect temperature data in a certain region
Raster data	recording observation data at fixed locations and fixed time	converting a raster to a finer or coarser resolution	estimate traffic volume using data from nearby sensors/cameras

9 of 79

Mining Spatial-Temporal Data: Data Models

Data Models for Spatial Data

Object model: points, lines, polygons (e.g., point ⇒ vehicle, line ⇒ road)
Field model: spatial information as a function (e.g., temperature in an area)
Spatial network model: graphs for relationship (e.g., road network)

Data Models for Temporal Data

Temporal snapshot model: multiple spatial layers associated with time.
Temporal change model: a spatial theme using a start time and the incremental changes (e.g., moving of a vehicle with initial location, speed)
Event/process model: multiple events and processes over time

10 of 79

Mining Rich Data Types

Mining Text Data
Mining Spatial-Temporal Data
Mining Graph and Networks

11 of 79

Mining Graph and Networks

Graph/Networks: consist of (1) a set of nodes and (2) a collection of edges, to model complex systems, e.g., online social platform
Categorization

attributed vs. plain graphs:

node/edge attributes exist or not.

dynamic vs. static graphs:

graph topology/attribute change or not.

homogeneous vs. heterogeneous graphs:

nodes/edges are of the same type or not.

Homogeneous vs. heterogeneous networks

12 of 79

Mining Graph and Networks: Problems

Graph Modeling: to study and simulate the generation mechanisms.

Motivation: (1) understanding how graphs are created and formed, (2) helping protect user privacy by anonymizing social networks.
Graph models:

Erdős–Rényi model: assumes certain probability distributions.
realistic graph generators: mimic one or more properties of real-world graphs, e.g., preferential attachment based (Barabási–Albert model).
deep generative models: mimic the real graphs with neural networks.

Challenge: heterogeneous networks

13 of 79

Mining Graph and Networks: Problems

Heterogeneous Network Mining: capturing rich semantics

Core concept: meta-path represents a sequence of relations, e.g., author-paper-author ⇒ co-authorship relation.
Applications of meta-path

Similarity search: improving ranking accuracy
Node classification: similarity matrices can be used to infer node labels
Clustering: representing different semantics
heterogeneous network embedding: constructing neighborhood

Challenge: automatically inferring structure and semantics from input data

14 of 79

Mining Graph and Networks: Problems

Knowledge Graph Mining

Knowledge graph (KG): a directed heterogeneous graph to link concepts and entities through human-interpretable semantics, e.g., DBPedia, Google KG.
KG construction methods: labor intensive, relying on pre-specified ontology

Extractive construction: information extraction, e.g., semantic labeling.
Language model-based
Representation learning-based: link prediction, entity resolution

Challenges: (1) automatic knowledge graph construction and dynamic update, (2) medical AI, conversational AI, and academic search engine

15 of 79

Summary: Mining Rich Data Types

Mining Text Data

Major tasks: information extraction, taxonomy construction, text clustering/classification/summarization
Techniques: symbol-based and distributed representation (e.g., word2vec)

Mining Spatial-Temporal Data

Types: event, trajectory, point reference data, raster data
Models: object/field/spatial network and snapshot/change/event-process

Mining Graph and Networks

Problems: graph modeling, heterogeneous graph mining, knowledge graph

16 of 79

Chapter 12. Data Mining Trends and Research Frontiers

Mining Rich Data Types
Data Mining Applications
Data Mining Methodologies and Systems
Data Mining, People and Society
Summary

17 of 79

Data Mining Applications

Data Mining for Sentiment and Opinion
Truth Discovery and Misinformation Identification
Information and Disease Propagation
Productivity and Team Science

18 of 79

Sentiment and Opinion

Sentiment: a view of attitude/emotion toward a situation or event.
Opinion: a view or judgement formed about something.
Sentiment and opinion are heavily correlated.

Note: many opinions may express sentiment, but some do not. (e.g., “I think she will move to Canada after graduation.”)

Subtle difference between sentiment and opinion:

Sentiment sentence example: “I find data mining highly interesting.”
Opinion sentence example: “I believe data mining is promising and useful.”
The subtle difference is that one may share the same sentiment expressed in the example of sentiment sentence and agree or disagree with the opinion in the second sentence.

19 of 79

Sentiment Analysis and Opinion Mining Techniques

Goal: understand sentiment expressed in text.
Techniques being used:

Natural language processing methods.
Machine learning and data mining methods.
Psychology and social sciences.

Technical directions being explored:

Analysis at different levels: Document-level, sentence-level, aspect-level.
Sentiment classification based on lexicons.
Two types of text content: stand-alone and online conversations.
Mining intent: intent may imply sentiments and express opinions.
Opinions spam detection: acquire data for social goodness.

20 of 79

Sentiment Analysis and Opinion Mining Applications

Understand advantages/disadvantages of targets:

Sentiment analysis/opinion mining over product reviews.
Help customer make decisions; Help platform optimize supply chain; Help producers and manufacturers improve products, etc.

Outcome can be consumed by other AI agents:

Example: sentiment from news trading signals for stock markets.
Challenges: highly efficient/accurate/robust system.

Monitoring and government administration:

Fight against cyber-violence, terrorism and racism.
Challenges: fairness and privacy issues.

21 of 79

Data Mining Applications

Data Mining for Sentiment and Opinion
Truth Discovery and Misinformation Identification
Information and Disease Propagation
Productivity and Team Science

22 of 79

Truth Discovery

Goal: assess and choose the actual true value for a data item when different sources provide conflicting information on it.
Single-truth discovery:

Only one true value is allowed for data item.

Multi-truth discovery:

Multiple true values are allowed for data items.

Challenges:

Source trustworthiness and claim reliability.
Limited amount of labeled data source: semi-supervised/unsupervised.
Detect copy behaviors of false values.
Evaluation: based on data distribution and treat those deviated substantially from norm as wrong answers.

23 of 79

Identification of Misinformation

Goal: detect false, inaccurate, or misleading information from data source.
Potential solutions:

Explore mutual enhancement between source trustworthiness and claim credibility (similar idea from truth discovery).
Methods should distinguish misinformation with additional measures:

Building up information literacy and media literacy through education.
Using commonsense knowledge and open-mindedness.
Developing critical thinking.

Determine misinformation in the final stage based on:

Individual’s mental model and worldview beliefs.
Consideration of repetition of misinformation.
Time-lag/relative coherency between misinformation and corrective information.

24 of 79

Data Mining Applications

Data Mining for Sentiment and Opinion
Truth Discovery and Misinformation Identification
Information and Disease Propagation
Productivity and Team Science

25 of 79

Information and Disease Propagation

Data mining research problems:

Prediction problems of propagation.
Optimization problems of propagation.

Applications:

Social media: predict which piece of information is likely to go viral; detect the rumor source who started a misinformation campaign; neutralize the propagation of misinformation.
Computational epidemiology: determine critical network condition (e.g., epidemic threshold) under which an epidemic is likely to happen.

26 of 79

Information and Disease Propagation

Prediction problems of propagation:

Classification and regression: whether information being popular/popular scores.
Prediction around publication: before/after information is published.
Prediction at different granularities: individual/groups of information.

Optimization problems of propagation (in a social network perspective):

Influence maximization: choose a set of initial nodes to maximize the number of infected nodes.
Source localization: identify source of a cascade over network with partial observations.
Activity shaping: steer user’s activity.
Graph connectivity optimization: manipulate the graph topology to affect the information propagation results.

27 of 79

Information and Disease Propagation

Models for information diffusion problems:

Feature based model:

Idea: utilize supervised models for information diffusion prediction.
Features used: temporal, local and global structure, user and item, etc.

Generative model:

Idea: take information diffusion as event sequences in the continuous temporal domain, and formulate as statistical generative approaches.
Methods: Poisson Process, Survival Analysis, Hawkes Process, Epidemic Model.

Deep learning based model:

Idea: no assumptions on the information diffusion process.
Methods: multi-modal data with various neural modules.

Computational epidemiology models: SIS, SIR and SEIR and their variants.

28 of 79

Data Mining Applications

Data Mining for Sentiment and Opinion
Truth Discovery and Misinformation Identification
Information and Disease Propagation
Productivity and Team Science

29 of 79

Productivity and Team Science

Team components: (1) team leader and members, (2) the environment in which team members collaborate, (3) the task on which the team works.
Data mining research questions:

Team performance characterization: how to reveal key characteristic patterns that differentiate a high-performing team from a struggling one?
Team performance prediction: how to forecast the performance of the team before or soon after the start of the task?
Team performance optimization: how to further enhance the team performance by adjusting the team composition?
Team performance explanation: how to interpret the team performance prediction and optimization results intuitively?

30 of 79

Productivity and Team Science

Team performance characterization:

Challenges:

Team members possess different type of skills and social connectivity.
The nature of tasks varies depends on the application scenarios.
Environments are large, noisy, incomplete and volatile to dynamics.

Treat as outcome of collective intelligence from both virtual teams and teams with face-to-face interactions.

Team performance prediction:

Challenges :

Identify crucial features.
Model the correlation between features and the team performance.
Encode the dynamics of team evolution.

Predict the impact of contents that team produces at a finer granularity.

31 of 79

Productivity and Team Science

Team performance optimization:

Optimize the team composition to maximize the performance.
Aspects considered by the replacement algorithms:

Skills of the team members.
Network connectivity (the environment).

Generalizations: team shrinkage, team expansion, team conflict resolution, real-time team optimization, etc.

Team performance explanation:

Limited literature exists for intuitive explanations of team performance/algorithms.
Use influence function to identify key elements in teams for interpretation.
Interpret from multiple levels.

32 of 79

Summary: Data Mining Applications

Data Mining for Sentiment and Opinion

detect sentiment in texts with techniques from NLP, ML/DM, psychology and social science.

Truth Discovery and Misinformation Identification

determine true values for data items.
determine false/inaccurate and misleading values for data items.

Information and Disease Propagation

prediction problem and optimization problem.

Productivity and Team Science

team performance characterization, prediction, optimizaion and explanation.

33 of 79

Chapter 12. Data Mining Trends and Research Frontiers

Mining Rich Data Types
Data Mining Applications
Data Mining Methodologies and Systems
Data Mining, People and Society
Summary

34 of 79

Data Mining Methodologies and Systems

Structuring Unstructured Data for Knowledge Mining: A Data-Driven Approach
Data Augmentation
From Correlation to Causality
Network as a Context
Auto-ML: Methods and Systems

35 of 79

Taxonomy construction and refinement

Taxonomy

Organizing important concepts into semantically rich structures
Playing an essential role at organizing massive unstructured text data into relatively organized structures

Suitable taxonomy properties

multi-faceted
corpus- and application-dependent

Taxonomy generation and expansion

Weakly or distantly supervised methods

Set expansion
Embedding-based hierarchical clustering
Adding new emerging terms or deleting old, obsolete ones

36 of 79

Weakly supervised text classification

Text: massive, diverse and in multiple granularities
Weakly supervised text classification

Idea #1:

Generate pseudo documents by Embedding methods
Train neural networks based on the pseudo documents and unlabeled text data

Idea #2:

Generate class-distinctive keywords or phrases by category-guided embedding

Idea #3:

Pre-trained language model to generate class-distinctive keywords

37 of 79

Fine-grained information extraction

Fine-grained entity recognition

Extract the concrete role that an entity plays in a particular context

Identification of the context of an entity

Generate entity mention candidates and relation phrases
Find the most appropriate fine-grained types for the entities to be examined

Methods

Distant supervision based methods
Embedding based methods
Taxonomy-guided supervision and pre-trained language model

38 of 79

Knowledge graph/information network construction

Knowledge graphs

A set of entities
A set of (possibly labeled) edges linking among entities

Construction of knowledge graph

Use distant or weak supervision
Taxonomy-guided text classification
Problems of building knowledge graph

An entity can be associated with different attributes/value
Easy to cause confusion if different associations are merged into a single knowledge graph

Solutions

Construct local knowledge graphs
Use such local knowledge graphs under similar conditions

39 of 79

Data Mining Methodologies and Systems

Structuring Unstructured Data for Knowledge Mining: A Data-Driven Approach
Data Augmentation
From Correlation to Causality
Network as a Context
Auto-ML: Methods and Systems

40 of 79

Data Augmentation Details

Why do we need data augmentation?

A high-performing data mining model often requires a great amount of labelled data samples for training
However, for many domains, we might be only provided by a limited number of labelled data
Training a data mining model with scarce training data, which could lead to the overfitting issue

Basic Methods: Flipping, rotation, equalization and so on
Augmentation Policy Learning: Combine multiple basic augmentations
GANs-Based Methods: Training a GAN model to generate new data
Adversarial Training: Mixing the trickily-perturbed samples into the training samples

41 of 79

Data Mining Methodologies and Systems

Structuring Unstructured Data for Knowledge Mining: A Data-Driven Approach
Data Augmentation
From Correlation to Causality
Network as a Context
Auto-ML: Methods and Systems

42 of 79

From Correlation to Causality

Correlation analysis

Study the statistical associations between different observed variables

Causal analysis

Causality relation between observed variables

Causality analysis tasks

Causality discovery

constraint-based algorithms, score-based algorithms and functional causal models

Causality inference

Regression adjustment, propensity score methods and covariate balancing methods

43 of 79

From Correlation to Causality

Causality analysis benefits

Enhance the interpretability of black box deep learning models
Help avoid unfair results
Improve the robustness of models

The taxonomy of causality analysis

44 of 79

Data Mining Methodologies and Systems

Structuring Unstructured Data for Knowledge Mining: A Data-Driven Approach
Data Augmentation
From Correlation to Causality
Network as a Context
Auto-ML: Methods and Systems

45 of 79

Network as a Context

Network of X

Each node represents an entity, a data set or a data mining model

Examples: Network of time series

The key idea for mining network of time series data

Model each time series by either traditional signal processing approaches or deep neural networks
Leverage the contextual network to regularize different time series models

46 of 79

Network as a Context

Network of X

Each node represents an entity, a data set or a data mining model

Examples: Network of networks

Each X (i.e., node) itself of the contextual network represents another domain-specific network
The main advantages of the network of networks model

Explicitly encodes the hierarchical structure
The contextual network provides additional regularization

47 of 79

Network as a Context

Network of X

Each node represents an entity, a data set or a data mining model

Examples: Network of data mining models

Each X itself is a data mining model
Different performance prediction models ‘borrow’ data from each other
Mutually boost each other’s prediction performance

Future directions

The contextual network construction
Can be applied to more applications, e.g., knowledge graphs, heterogeneous networks
Integrate other data mining problems, e.g., adversarial learning, fairness, explainable learning.

48 of 79

Data Mining Methodologies and Systems

Structuring Unstructured Data for Knowledge Mining: A Data-Driven Approach
Data Augmentation
From Correlation to Causality
Network as a Context
Auto-ML: Methods and Systems

49 of 79

Auto-ML: Methods and Systems

Why do we need Auto-ML

Different data mining and machine learning models have been widely used
Different task has different settings
Building and training appropriates mining models for each task could be time consuming
Domain experts might not be familiar with the mining models

Auto-ML classification

Automated Feature Engineering (Auto-FE)

Automatically detects the most representative and informative features

Automated Model and Hyperparameter Tuning (Auto-MHT)

Automatically builds machine learning models and tunes the hyper-parameters

50 of 79

Auto-ML: Methods and Systems

Key challenges for Auto-ML

The lack of authoritative benchmarks for Auto-FE and Auto-MHT
The efficiency of Auto-ML
How to incorporate human knowledge or experiences into the AutoML to find suitable sub-space of the entire search space
The interpretability of Auto-ML
The scope of applications

Taxonomy of Auto-ML

51 of 79

Summary: Data Mining Methodologies and Systems

Structuring Unstructured Data for Knowledge Mining: A Data-Driven Approach

Transform unstructured text-rich data into organized, relatively structured data

Data Augmentation

We might be only provided by a limited number of labelled data
Data augmentation can prevent overfitting

From Correlation to Causality

Focus on the causality relation between observed variables

52 of 79

Summary: Data Mining Methodologies and Systems

Network as a Context

Provided a powerful context that links different types of data from different sources with different data mining algorithms
Auto-ML: Methods and Systems

Allow non-experts to make use of machine learning models and techniques without requiring them to become experts in machine learning.

53 of 79

Chapter 12. Data Mining Trends and Research Frontiers

Mining Rich Data Types
Data Mining Applications
Data Mining Methodologies and Systems
Data Mining, People and Society
Summary

54 of 79

Data Mining, People and Society

Privacy-Preserving Data Mining
Human-Algorithm Interaction
Mining beyond Maximizing Accuracy: Fairness, Interpretability, and Robustness
Data Mining for Social Good
Summary

55 of 79

Privacy-Preserving Data Mining

Privacy leakage between users and data owners

Example: re-identify users based on anonymous reviews on products.

Anonymize data for privacy protection

k-anonymity (P. Samarat et al., 1998): generalize the data so that it is identical to at least k-1 other records on the identifying attribute.

Differential privacy (A. Machanavajjhala et al. 2006)

Describe patterns of groups without sharing individual information.
Global sensitivity: for a function f and a pair of neighboring datasets D₁ and D₂, GS_f is the maximum difference between function value f(D₁) and f(D₂).

56 of 79

Privacy-Preserving Data Mining

Privacy leakage between data owners

Example: multiple companies collaboratively conduct data mining but refuse to share user information.
Solution: federated learning and analytics.

Federated learning

Run local computation over each data owners’ devices and aggregate results without leaking detailed data from individual data owners.
Example: federated K-Means.
Horizontal and vertical federated learning

Horizontal: data owned by each data owners follows the same schema.
Vertical: different data owners have the data on only some attributes/objects.

57 of 79

Data Mining, People and Society

Privacy-Preserving Data Mining
Human-Algorithm Interaction
Mining beyond Maximizing Accuracy: Fairness, Interpretability, and Robustness
Data Mining for Social Good

58 of 79

Human-Algorithm Interaction

Human-algorithm interaction

Goal: optimize the collaboration between human intelligence and algorithm, and consequently enhance the system performance and user experience.

Crowdsourcing

Goal: harness the human intelligence to resolve difficult problems.
Applications: label generation, model evaluation, evaluating prediction interpretability, debug AI systems, etc.

Human-in-the-loop

Goal: leveraging human intervention for improving algorithms.
Applications: high-quality label generation for supervised learning, enhancing model explainability with background knowledge, model evaluation, etc.

59 of 79

Human-Algorithm Interaction

Machine-in-the-loop

Goal: improving humans’ query strategy by algorithms.
Applications: creative writing and drawing, achieving comparable model performance which reducing the demand for labelled data

Human-machine-teaming

Goal: fostering the effective teamwork between humans and algorithms.
Applications:

60 of 79

Data Mining, People and Society

Privacy-Preserving Data Mining
Human-Algorithm Interaction
Mining beyond Maximizing Accuracy: Fairness, Interpretability, and Robustness
Data Mining for Social Good

61 of 79

Mining beyond Maximizing Accuracy:

Fairness, Interpretability, and Robustness

Algorithm fairness: definitions

Group fairness

Demographic parity: predictions are independent of sensitive attributes
Equalized odds: conditional probability of the same prediction results given the label should be equal for different sensitive attributes.
Predictive rate parity: conditional probability of label given the prediction results should be equal for different sensitive attributes.

Individual fairness: preserve individual similarities, i.e., similar individuals should be treated similarly.
Counterfactual fairness: correct predictions of a label variable that are unfairly altered by an individual’s sensitive attribute.

62 of 79

Mining beyond Maximizing Accuracy:

Fairness, Interpretability, and Robustness

Algorithm fairness: strategies

Pre-processing

Learn representations that removes the impact of sensitive attributes.

Optimization at training

Use regularizers or constraints to achieve trade-off between mitigating the bias and retaining the original mining accuracy.

Post-processing

Find a proper threshold for each group to achieve certain types of fairness

63 of 79

Mining beyond Maximizing Accuracy:

Fairness, Interpretability, and Robustness

Interpretability

Goal: explain results given by ‘black boxes’.
Visualize summary statistics for each feature as interpretation.
Explore self-interpretable components of the models (e.g., weights in the linear model, structure of decision trees)
Identifying key data points for interpretation.

Robustness

Goal: ensuring consistency between test results and training results with unintentionally added noises or intentional adversarial attacks.
Reactive methods: detect adversarial examples after the model is built.
Proactive methods: make mining models more robust to attacks

64 of 79

Data Mining, People and Society

Privacy-Preserving Data Mining
Human-Algorithm Interaction
Mining beyond Maximizing Accuracy: Fairness, Interpretability, and Robustness
Data Mining for Social Good

65 of 79

Data Mining for Social Good

Data mining applications

Challenges

Data scarcity: lack of large-scale data
Evaluation: standard evaluation metrics may be insufficient in real-world scenarios
Human-in-the-loop: insufficient use of domain knowledge from experts
Sustainable deployment

66 of 79

Summary: Data Mining, People and Society

Privacy-Preserving Data Mining

Privacy between user and data owner: differential privacy
Privacy between data owners: federated learning

Human-Algorithm Interaction

Crowdsourcing
Human-in-the-Loop
Machine-in-the-Loop
Human-Machine-teaming

Beyond Accuracy

Fairness: group fairness, individual fairness, and counterfactual fairness
Interpretability
Robustness

Data Mining for Social Good

67 of 79

References

Pierangela Samarati and Latanya Sweeney. Protecting Privacy when Disclosing Information: k-Anonymity and its Enforcement through Generalization and Suppression. Technical report, SRI International, 1998.
A. Machanavajjhala, J. Gehrke, D. Kifer, and M. Venkitasubramaniam. L-diversity: privacy beyond k-anonymity. In Proc. 2006 Int. Conf. Data Engineering (ICDE’06), pages 24–24, Atlanta, GA, USA, Apr. 2006.
Vikas C. Raykar, Shipeng Yu, Linda H. Zhao, Gerardo Hermosillo Valadez, Charles Florin, Luca Bogoni, and Linda Moy. Learning from crowds. Journal of Machine Learning Research, 11(43):1297– 1322, 2010.

68 of 79

Chapter 12. Data Mining Trends and Research Frontiers

Mining Rich Data Types
Data Mining Applications
Data Mining Methodologies and Systems
Data Mining, People and Society
Summary

69 of 79

Summary: Data Mining Trends and Research Frontiers

Mining Rich Data Types

Text data
Spatial-Temporal data
Graph and Networks

Data Mining Methodologies and Systems

Structure Unstructured Data for Knowledge Mining
Data Augmentation
From Correlation to Causality
Network as a Context
Auto-ML

70 of 79

Summary: Data Mining Trends and Research Frontiers

Data Mining Applications

Data Mining for Sentiment and Opinion
Truth Discovery and Misinformation Identification
Information and Disease Propagation
Productivity and Team Science

Data Mining, People and Society

Privacy-Preserving Data Mining
Human-Algorithm Interaction
Mining beyond Maximizing Accuracy: Fairness, Interpretability, and Robustness
Data Mining for Social Good

71 of 79

References (Mining Rich Data Types)

T. Mikolov, I. Sutskever, K. Chen, G. S. Corrado, and J. Dean. Distributed representations of words and phrases and their compositionality. NIPS'13
J. Pennington, R. Socher, and C. D. Manning. GloVe: Global vectors for word representation. EMNLP'14
Y. Meng, J. Huang, G. Wang, C. Zhang, H. Zhuang, L. M. Kaplan, and J. Han. Spherical text embedding. NeurIPS'19
M. Nickel and D. Kiela. Poincaré embeddings for learning hierarchical representations. NIPS’17
J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. NAACL-HLT'19

72 of 79

References (Mining Rich Data Types)

G. Atluri, A. Karpatne, and V. Kumar. Spatiotemporal data mining: A survey of problems and methods. ACM Comput. Surv., August 2018
S. Shekhar, R. Vatsavai, and M. Celik. Spatial and spatiotemporal data mining: Recent advances. Next Generation of Data Mining, 2008.
P. W. Holland, K. B. Laskey, and S. Leinhardt. Stochastic blockmodels: First steps. Social networks, 5, 1983.
J. Leskovec, D. Chakrabarti, J. Kleinberg, C. Faloutsos, and Z. Ghahramani. Kronecker graphs: an approach to modeling networks. JMLR, 11, 2010.
Y. Dong, N. V. Chawla, and A. Swami. metapath2vec: Scalable representation learning for heterogeneous networks. KDD’17

73 of 79

References (Data Mining Methodologies and Systems)

J. G. Shanahan, Y. Qu, and J. Wiebe, editors. Computing Attitude and Affect in Text: Theory and Applications. Springer, 2006.
B. Liu. Sentiment Analysis: Mining Opinions, Sentiments, and Emotions. Cambridge University Press, 2020.
Y. Li, J. Gao, C.i Meng, Q. Li, L. Su, B. Zhao, W. Fan, and J. Han. A survey on truth discovery. SIGKDD, 2015.
L. Wu, F. Morstatter, K. M. Carley, and H. Liu. Misinformation in social media: Definition, manipulation, and detection. SIGKDD’19.
Z. Chen, H. Tong, and L. Ying. Inferring full diffusion history from partial timestamps. TKDE’19.

74 of 79

References References (Data Mining Methodologies and Systems)

Y. Yang and J. Pei. Influence analysis in evolving networks: A survey. TKDE’19.
S. F Ardabili, A. Mosavi, P. Ghamisi, F. Ferdinand, A. R Varkonyi-Koczy, U. Reuter, T. Rabczuk, and P. M Atkinson. Covid-19 outbreak prediction with machine learning. Algorithms, 2020.
J. Hackman and N. Katz. Group Behavior and Performance, 2010.
L. Li and H. Tong. Computational Approaches to the Network Science of Teams. Cambridge University Press, 2020.

75 of 79

References (Data Mining Applications)

Jingbo Shang, Jialu Liu, Meng Jiang, X. Ren, Clare R. Voss, and Jiawei Han. Automated phrase mining from massive text corpora. IEEE Transactions on Knowledge and Data Engineering, 2018.
Chi Wang and Jiawei Han. Mining Latent Entity Structures. Morgan & Claypool, 2015.
Xiang Ren, Ahmed El-Kishky, Chi Wang, Fangbo Tao, Clare R. Voss, Heng Ji, and Jiawei Han. ClusType: Effective entity recognition and typing by relation phrase-based clustering. KDD, 2015.
Jiaxin Huang, Yiqing Xie, Yu Meng, Jiaming Shen, Yunyi Zhang, and Jiawei Han. Guiding corpus-based set expansion by auxiliary sets generation and co-expansion. The Web Conf. (WWW’20), 2020.

76 of 79

References (Data Mining Applications)

Ekin Dogus Cubuk, Barret Zoph, Dandelion Man´e, Vijay Vasudevan, and Quoc V. Le. Autoaugment: Learning augmentation policies from data. CoRR, 2018.
Siyu Shao, Pu Wang, and Ruqiang Yan. Generative adversarial networks for data augmentation in machine fault diagnosis. Comput, 2019.
Antreas Antoniou, Amos J. Storkey, and Harrison Edwards. Data augmentation generative adversarial networks. CoRR, 2017.
Peter Spirtes, Clark N Glymour, Richard Scheines, and David Heckerman. Causation, prediction, and search. MIT press, 2000.
David Maxwell Chickering. Optimal structure identification with greedy search. Journal of machine learning research, 2002.

77 of 79

References (Data Mining Applications)

Keisuke Hirano, Guido W Imbens, and Geert Ridder. Efficient estimation of average treatment effects using the estimated propensity score. Econometrica, 2003.
Christos Louizos, Uri Shalit, Joris Mooij, David Sontag, Richard Zemel, and Max Welling. Causal effect inference with deep latent variable models. arXiv, 2017.
Yongjie Cai, Hanghang Tong, Wei Fan, and Ping Ji. Fast mining of a network of coevolving time series. In Proc. 2015 SIAM Int. Conf. Data Mining (SDM’15), 2015.
Yongjie Cai, Hanghang Tong, Wei Fan, Ping Ji, and Qing He. Facets: Fast comprehensive mining of coevolving high-order time series. KDD, 2015.

78 of 79

References (Data Mining, People and Society)

Hannes Heikinheimo and Antti Ukkonen. The crowd-median algorithm. In Proc. 2013 Conf. Human Computation and Crowdsourcing (HCOMP’13), Palm Springs, CA, USA, Nov. 2013.
Flavio P Calmon, Dennis Wei, Bhanukiran Vinzamuri, Karthikeyan Natesan Ramamurthy, and Kush R Varshney. Optimized pre-processing for discrimination prevention. In Proc. 2017 Conf. Neural Information Processing Systems (NIP’17), pages 3995–4004, Long Beach, CA, USA, Dec. 2017.
Kui Ren, Tianhang Zheng, Zhan Qin, and Xue Liu. Adversarial attacks and defenses in deep learning. Engineering, 6(3):346–360, 2020.
Christoph Molnar. Interpretable machine learning. Lulu. com, 2020.

79 of 79

References (Data Mining, People and Society)

Wojciech Samek, Thomas Wiegand, and Klaus-Robert Mu ̈ller. Explainable artificial intelligence: Understanding, visualizing and interpreting deep learning models. arXiv preprint arXiv:1708.08296, 2017.
Jacob Abernethy, Alex Chojnacki, Arya Farahi, Eric Schwartz, and Jared Webb. Activeremediation: The search for lead pipes in flint, michigan. In Proc. 2018 ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining (KDD’18), pages 5–14, London, UK, Aug. 2018.
Kaixiang Lin, Renyu Zhao, Zhe Xu, and Jiayu Zhou. Efficient large-scale fleet management via multi-agent deep reinforcement learning. In Proc. 2018 ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining (KDD’18), pages 1774–1783, London, UK, Aug. 2018.