In the last years, data lakes are emerging as an e ective and e cient support for information and knowledge extraction from a huge amount of highly heterogeneous and quickly changing data sources. Data lake management requires the de nition of new techniques, very di erent from the ones adopted for data warehouses in the past. One of the main issues to address in this scenario consists in the extraction of thematic views from the (very heterogeneous and generally unstructured) data sources of a data lake. In this paper, we propose a new network-based model to uniformly represent structured, semi-structured and unstructured sources of a data lake. Then, we present a new approach to, at least partially, \structure" unstructured data. Finally, we de ne a technique to extract thematic views from the sources of a data lake, based on similarity and other semantic relations among the metadata of data sources.
In the last years, data lakes are emerging as an e ective and e cient answer
to the problem of extracting information and knowledge from a huge amount
of highly heterogeneous and quickly changing data sources [
In this paper, we aim at providing a contribution in this setting. Indeed, we
propose a supervised approach to extracting thematic views from highly
heterogeneous sources of a data lake. Our approach represents all the data lake sources
by means of a suitable network. Indeed, networks are very exible structures that
allow the modeling of almost all phenomena that researchers aim at investigating
[
This paper is organized as follows: Section 2 illustrates related literature. In Section 3, we present the proposed approach. In particular, rst we describe a unifying model for data lake representation; then, we present our approach to partially structuring unstructured sources; nally, we discuss the two steps of our approach for thematic view extraction. In Section 4, we present our example case, whereas, in Section 5, we draw our conclusions and discuss future work. 2
The new data lake scenario is characterized by several peculiarities that make
it very di erent from the data warehouse paradigm. Hence, it is necessary to
adapt (if possible) old algorithms conceived for data warehouses or to de ne
new approaches capable of handling and taking advantage of the speci cities of
this new paradigm. However, most approaches proposed in the literature for data
integration, query answering and view extraction do not completely t the data
lake paradigm. For instance, [
DBpedia: http://dbpedia.org databases; it requires the presence of a static context and the pre-computation of some queries.
Another family of approaches exploits materialized views to perform tree
pattern querying [
Finally, semantic-based approaches have long been used to drive data
integration in databases and data warehouses. More recently, in the context of big
data, formal semantics has been speci cally exploited to address issues
concerning data variety/heterogeneity, data inconsistency and data quality in such a
way as to increase understandability [
In this section, we illustrate our network-based model to represent and handle a data lake, which we will use in the rest of this paper. In our model, a data lake DL is represented as a set of m data sources: DL = fD1; D2; ; Dmg. A data source Dk 2 DL is provided with a rich set Mk of metadata. We denote with MDL the repository of the metadata of all the data sources of DL: MDL = fM1; M2; : : : ; Mmg.
According to [
Syntax: http://www.w3.org/TR/2004/ is commonly provided by the source catalogue. (ii) MkO represents operational metadata. It includes the source and target locations of the corresponding data, the associated le size, the number of their records, and so on. Usually, it is autoB matically generated by the technical framework handling the data lake. (iii) Mk indicates business metadata. It comprises the business names and descriptions assigned to data elds. It also covers business rules, which can become integrity constraints for the corresponding data source.
In this paper, we consider only business metadata. Indeed, they denote, at the intensional level, the information content stored in the data lake sources and are those of interest for supporting the extraction of thematic views from a data lake, which is our ultimate goal. In order to represent MkB, our model adopts a notation typical of XML, JSON and many other semi-structured models. According to this notation, Objk indicates the set of all the objects stored in MkB. It consists of the union of three subsets: Objk = Attk [ Smpk [ Cmpk. Here: (i) Attk indicates the set of the attributes of MkB; (ii) Smpk represents the set of the simple elements of MkB; (iii) Cmpk denotes the set of the complex elements of MkB. In this context, the meaning of the terms \attribute", \simple element" and \complex element" is the one typical of semi-structured data models.
MkB can be also represented as a graph: MkB = hNk; Aki. Nk is the set of the nodes of MkB. There exists a node nkj 2 Nk for each object okj 2 Objk. According to the structure of Objk, Nk consists of the union of three subsets: Nk = NkAtt [ NkSmp [ NkCmp. Here, NkAtt (resp., NkSmp, NkCmp) indicates the set of the nodes corresponding to Attk (resp., Smpk, Cmpk). There is a oneto-one correspondence between a node of Nk and an object of Objk. Therefore, in the following, we will use the two terms interchangeably. Let x be a complex element of MkB. Objkx indicates the set of the objects directly contained in x, whereas NkOxbj denotes the set of the corresponding nodes. Finally, let y be a simple element of MkB. Attky represents the set of the attributes of y, whereas NkAytt denotes the set of the corresponding nodes.
Ak indicates the set of the arcs of MkB. It consists of two subsets: Ak = A0k [ A00. Here: (i) A0k = f(nx; ny)jnx 2 N Cmp; ny 2 NnOxbj g, i.e., there is an k k arc from a complex element of MkB to each object directly contained in it. (ii) A0k0 = f(nx; ny)jnx 2 NkSmp; ny 2 NnAxttg, i.e., there is an arc from a simple element of MkB to each of its attributes. 3.2
Our network-based model for representing and handling a data lake is perfectly tted for representing and managing semi-structured data because it has been designed having XML and JSON in mind. Clearly, it is su ciently powerful to represent structured data. The highest di culty regards unstructured data because it is worth avoiding a at representation consisting of a simple element for each keyword provided to denote the source content. As a matter of fact, this kind of representation would make the reconciliation, and the next integration, of an unstructured source with the other (semi-structured and structured) ones of the data lake very di cult. Therefore, it is necessary to (at least partially) \structure" unstructured data.
Our approach to addressing this issue consists of four phases, namely: (i) creation of nodes; (ii) derivation and management of part-of relationships; (iii) derivation of lexical and string similarities; (iv) management of lexical and string similarities.
Phase 1. During this phase, our approach creates a complex element representing the source as a whole, and a simple element for each keyword. Furthermore, it adds an arc from the source to each of the simple elements. Initially, there is no arc between two simple elements. To determine the arcs to add, the next phases are necessary.
Phase 2. During this phase, our approach adds an arc from the node nk1 ,
corresponding to the keyword k1, to the node nk2 , corresponding to the keyword
k2, if k2 is registered as a lemma of k1 in a suitable thesaurus. Taking the
current trends into account, this thesaurus should be a multimedia one; for this
purpose, in our experiments, we have adopted BabelNet [
Phase 3. During this phase, our approach starts by deriving lexical
similarities. In particular, it states that there exists a similarity between the nodes nk1 ,
corresponding to the keyword k1, and nk2 , corresponding to the keyword k2, if k1
and k2 have at least one common lemma in a suitable thesaurus. Also in this case,
we have adopted BabelNet. After having found lexical similarities, our approach
derives string similarities and states that there exists a similarity between nk1
and nk2 if the string similarity degree kd(k1; k2), computed by applying a suitable
string similarity metric on k1 and k2, is \su ciently high" (see below). We have
chosen N-Grams [
Phase 4. During this phase, our approach aims at managing the similarities
found during Phase 3. In particular, if there exists a (lexical and/or string)
simIn this paper, we use the term \lemma" according to the meaning it has in
BabelNet [
Our approach to extracting thematic views operates on a data lake DL whose data sources are represented by means of the model described in Section 3.1. It is called \supervised" because it requires the user to specify the set of topics T = fT1; T2; ; Tlg, which should be present in the thematic view(s) to extract. It consists of two steps, the former mainly based on the structure of the sources at hand, the latter mainly focusing on the corresponding semantics. We describe these two steps in the next subsections.
Step 1. The rst step of our approach receives a data lake DL, a set of topics
T = fT1; T2; ; Tlg, representing the themes of interest for the user, and a
dictionary Syn of synonymies involving the objects stored in the sources of DL.
This dictionary could be a generic thesaurus, such as BabelNet [
In this step, the concept of ego network [
Let Ti be a topic of T . Let Obji = foi1 ; oi2 ; ; oiq g be the set of the objects synonymous of Ti in DL. Let Ni = fni1 ; ni2 ; ; niq g be the corresponding nodes. First, Step 1 constructs the ego networks Ei1 ; Ei2 ; ; Eiq having ni1 ; ni2 ; ; niq as the corresponding egos. Then, it merges all the egos into a unique node ni. In this way, it obtains a unique ego network Ei from Ei1 ; Ei2 ; ; Eiq . If a synonymy exists between two alters belonging to di erent ego networks, then these are merged into a unique node and the corresponding arcs linking them to the ego ni are merged into a unique arc. At the end of this task, we have a unique ego network Ei corresponding to Ti.
After having performed the previous task for each topic of T , we have a set E = fE1; E2; ; Elg of l ego networks. At this point, Step 1 nds all the synonymies of Syn involving objects of the ego networks of E and merges the corresponding nodes. After all the possible synonymies involving objects of the ego network of E have been considered and the corresponding nodes have been merged, a set V = fV1; ; Vgg, 1 g l, of networks representing potential views is obtained.
If g = 1, then it is possible to conclude that Step 1 has been capable of extracting a unique thematic view comprising all the topics required by the user. Otherwise, there exist more views each comprising some (but not all) of the topics of interest for the user. If g = 1, Step 2 is performed to make more precise and complete the unique view representing all the topics of T . If g > 1, Step 2 aims at nding other relationships, di erent from synonymies, among the objects of the views of V in such a way as to try to obtain a unique view embracing all the topics of interest for the user.
Step 2. This step starts by enriching each view Vi 2 V . For this purpose, it connects each of its elements to all the semantically related concepts taken from a reference knowledge repository.
In this work, we rely on DBpedia, one of the largest knowledge graphs in the Linked Data context, including more than 4.58 million entities in RDF. To this aim, rst each element of Vi (including its synonyms) is mapped to the corresponding entry in DBpedia. In many cases, such a mapping is already provided by BabelNet. Then, for each DBpedia entry, all the related concepts are retrieved. In DBpedia, knowledge is structured according to the Linked Data principles, i.e. as an RDF graph built by triples. Each triple hs(ubject); p(roperty); o(bject)i states that a subject s has a property p, whose value is an object o. Both subjects and properties are resources (i.e., nodes in DBpedia's knowledge graph), whereas objects may be either resources or literals (i.e., values of some primitive data types, such as strings or numbers). Each triple represents the minimal component of the knowledge graph. This last is built by merging triples together. Therefore, retrieving the related concepts for a given element x implies nding all the triples where x is either the subject or the object.
For each view Vi 2 V , the procedure to extend it consists of the following three substeps: 1. Mapping : for each node n 2 Vi, its corresponding DBpedia entry d is found. 2. Triple extraction: all the related triples hd; p; oi and hs; p; di, i.e., all the triples in which d is either the subject or the object, are retrieved. 3. View extension: for each retrieved triple hd; p; oi (resp., hs; p; di), Vi is extended by de ning a node for the object o (resp., s), if not already existing, linked to n through an arc labeled as p.
These three tasks are repeated for all the views of V . As previously pointed out, this enrichment procedure is particularly important if jV j > 1 because the new derived relationships could help to merge the thematic views that was not possible to merge during Step 1. In particular, let Vi 2 V and Vj 2 V be two views of V , and let Vi0 and Vj0 be the extended views corresponding to them. If there exist two nodes nih 2 Vi0 ad njk 2 Vj0 such that nih = njk , then they can be merged in one node; if this happens, Vi0 and Vj0 become connected. After all equal nodes of the views of V have been merged, all the views of V could be either merged in one view or not. In the former case, the process terminates with success. Otherwise, it is possible to conclude that no thematic views comprising Whenever this does not happen, the mapping can be automatically provided by the DBpedia Lookup Service (http://wiki.dbpedia.org/projects/dbpedia-lookup).
Here, two nodes are equal if the corresponding name coincide. all the topics speci ed by the user can be found. In this last case, our approach still returns the enriched views of V and leaves the user the choice to accept of reject them. 4
In this section, we present an example case aiming at illustrating the various tasks of our approach. Here, we consider: (i) a structured source, called Weather Conditions (W , in short), whose corresponding E/R schema is not reported for space limitations; (ii) two semi-structured sources, called Climate (C, in short) and Environment (E, in short), whose corresponding XML Schemas are not reported for space limitations; (iii) an unstructured source, called Environment Video (V , in short), consisting of a YouTube video and whose corresponding keywords are: garden, f lower, rain, save, earth, tips, recycle, aurora, planet, garbage, pollution, region, lif e, plastic, metropolis, environment, nature, wave, eco, weather, simple, f ineparticle, climate, ocean, environmentawareness, educational, reduce, power, bike.
By applying the approaches mentioned in Section 3.2, we obtain the corresponding representations in our network-based model, shown in Figure 1.
Assume, now, that a user speci es the following set T of topics of her interest: T = fOcean; Areag. First, our approach determines the terms (and, then, the objects) in the ve sources that are synonyms of Ocean and Area. As for Ocean, the only synonym present in the sources is Sea; as a consequence, Obj1 comprises the node Ocean of the source V (V:Ocean) and the node Sea of the source C (C:Sea). An analogous activity is performed for Area. At the end of this task we have that Obj1 = fV:Ocean; C:Seag and Obj2 = fW:P lace; C:P lace; V:Region; E:Locationg.
Step 1 of our approach proceeds by constructing the ego networks corresponding to the objects of Obj1 and Obj2. They are reported in Figure 2.
Now, consider the ego networks corresponding to V:Ocean and C:Sea. Our approach merges the two egos into a unique node. Then, it veri es whether further synonyms exist between the alters. Since none of these synonyms exists, it returns the ego network shown in Figure 3(a). The same task is performed to the ego networks corresponding to W:P lace, C:P lace, V:Region and E:Location. In particular, rst the four egos are merged. Then, synonyms between the alters W:City and C:City and the alters W:Altitude and C:Altitude are retrieved. Based on this, W:City and C:City are merged in one node, W:Altitude and C:Altitude in another node, the arcs linking the ego to W:City and C:City are merged in one arc and the ones linking the ego to W:Altitude and C:Altitude in another arc. In this way, the ego network shown in Figure 3(b) is returned. At this point, there are two ego networks, EOcean and EArea, each corresponding to one of the terms speci ed by the user.
Here and in the following, we use the notation S:o to indicate the object o of the source S.
Now, Step 1 veri es if there are any synonyms between a node of EOcean and a node of EArea. Since this does not happen, it terminates and returns the set V = fVOcean; VAreag, where VOcean (resp., VArea) coincides with EOcean (resp., EArea).
Fig. 2. Ego networks corresponding to V:Ocean, C:Sea, W:P lace, C:P lace, V:Region and E:Location.
At this point, Step 2 is executed. As shown in Figure 4, rst each term (synonyms included) is semantically aligned to the corresponding DBpedia entry (e.g., Ocean is linked to dbo:Sea, Area is linked to dbo:Location and dbo:Place, while Country to dbo:Country , respectively). After a single iteration, the triples hdbo:sea rdfs:range dbo:Seai and hdbo:sea rdfs:domain dbo:Placei are retrieved. They correspond to the DBpedia property sea, which relates a country to the sea from which it is lapped. Other connections can be found by moving to speci c instances of the mentioned resources. In this way, the following triples are retrieved: hinstance rdf:type dbo:Seai, hinstance rdf:type dbo:Locationi, hinstance rdf:type dbo:Placei, meaning that there are resource instances having types Sea, Location and Place simultaneously (e.g., dbr:Mediterranean Sea). Furthermore, a triple hinstance dbo:country dbo:Countryi can be retrieved, meaning that those instances being a Sea, a Location or a Place specify in which dbo:Country they are located through the dbo:country property. In this example case, Step 2 succeeded in merging the two views that Step 1 had maintained separated. 5
In this paper, we have presented a new network-based model to uniformly represent the structured, semi-structured and unstructured sources of a data lake. Then, we have proposed a new approach to, at least partially, \structuring" Pre xes dbo and dbr stand for http://dbpedia.org/ontology/ and http:// dbpedia.org/resource/ unstructured data. Finally, based on these two tools, we have de ned a new approach to extracting thematic views from the sources of a data lake consisting of two steps, based on ego networks (Step 1) and semantic relationships (Step 2). This paper is not to be intended as an ending point. Actually, it could be the starting point of a new family of approaches aiming at handling information systems in the new big data-oriented scenario. By proceeding in this direction, rst we plan to de ne an unsupervised approach to extracting thematic views from a data lake. Then, we plan to de ne new approaches to supporting a exible and lightweight querying of the sources of a data lake, as well as approaches to schema matching, schema mapping, data reconciliation and integration strongly oriented to data lakes based mainly on unstructured data sources.