Service Composition Based on Natural Language Requests

Abstract

Abstract-The easiest way for a user to express his needs regarding a desired service is to use natural language. The main issues come from the fact that the natural language is incomplete and ambiguous, while the service composition process should lead to valid services. In this paper we propose a natural language service assemblage method based on composition templates (patterns). The use of templates assures that the composition result is always valid. The proposed system, called NLSC (Natural Language Service Composer), was implemented on the top of a service-oriented middleware called WComp and tested in an intelligent home environment.

NLSC: Unrestricted Natural Language-Based Service Composition through Sentence Embeddings

“…In our study, we review natural language-based approaches for dynamic service composition. If we consider an user's natural language description at one end of the problem and services at the other end, then, we find that existing literature can be broadly categorized as approaches that a) apply restrictions on how the user expresses the goal using sentence templates and/or user utterances and then use structured parsing techniques to parse the sentences against service descriptions [5], [21]; b) construct semantic graphs that represent the service description [13] [28] [27] such that those could be matched with the natural language descriptions using a lexical database such as WordNet, that groups words based on their meanings, to calculate a conceptual distance metric between concepts [23] [10]; and c) match partiallyobservable natural language description using semantic web services such as OWL-S [24] [9]. Categorical limitations of existing approaches include, (i) complex linguistic processing that employs several NLP techniques: structured parsing, extracting parts-of-speech tokens, stop-word removal, spellchecking, stemming, and text segmentation, (ii) inclusion of lexical databases such as WordNet or domain-specific ontologies that represents domain lexicons, and (iii) a weaker concept representation and similarity score for semantic matching that does not account for sentence context.…”

NLSC: Unrestricted Natural Language-based Service Composition through Sentence Embeddings

“…In our study, we review natural language-based approaches for dynamic service composition. If we consider an user's natural language description at one end of the problem and services at the other end, then, we find that existing literature can be broadly categorized as approaches that a) apply restrictions on how the user expresses the goal using sentence templates and/or user utterances and then use structured parsing techniques to parse the sentences to match against service names and descriptions [4] [23]; b) construct semantic graphs that represent the service description [13] such that those could be matched with the natural language descriptions using a lexical database such as WordNet, that groups words based on their meanings, to calculate a conceptual distance metric between concepts at both ends, [26] [9]; and c) match partially-observable natural language description with that of the semantics of the service described using semantic web services such as OWL-S and VDL [27] [8]. Categorical limitations of existing approaches include, (i) complex linguistic processing that employs several NLP techniques: structured parsing, extracting parts-of-speech tokens, stop-word removal, spell-checking, stemming, and text segmentation, (ii) inclusion of lexical databases such as WordNet or domain-specific ontologies that represents domain lexicons, and (iii) a weaker concept representation and similarity score for semantic matching that does not account for sentence context.…”

“…In natural language-based service composition middleware, endusers interact instinctively with systems in natural language and expect the system to identify services that meet their goals. These kind of middleware can be broadly categorized as those that: (a) apply restrictions on how the user expresses the goal with sentence templates and then use structured parsing to match against service descriptions [4], [23]; (b) construct semantic graphs to represent service descriptions and match against a lexical database such as WordNet to compute concept similarity [9], [13], [26]; and (c) match partiallyobservable natural language request with semantics of service description expressed using semantic web services (OWL-S, VDL) [8], [27]. Limitations with these approaches include: (a) complex linguistic processing that requires additional natural language processing (NLP) techniques: structured parsing, extracting parts-of-speech, stop-word removal, spell-checking, stemming, and text segmentation; (b) inclusion of lexical databases such as WordNet or domain-specific ontologies; and (c) a weaker concept representation and similarity score for semantic matching that does not account for sentence context.…”

Towards Services Discovery Based on Service Goal Extraction and Recommendation

“…The paper mentions the important role played by modeling users' requirements in services discovery, which are represented as intentions using the MAP formalism. Cremene et al present an approach on automatic matching of Web services to user queries written in natural language [10]. Wang et al apply K-means algorithm to cluster both requirements and candidate services, which can effectively reduce the search space [11].…”