Abstract
Formal concept analysis with an incomplete context has received much attention recently, where an object is known to have one set of attributes and not have another set of attributes; for the rest of attributes, it is unknown if the object has or does not have them. This has led to a notion called partially-known formal concepts in a framework of three-way concept analysis with interval sets. The intent and/or extent of a partially-known concept may no longer be a set but an interval set. Depending on the set or interval set representation of the intent and extent, there are three different forms of partially-known formal concepts, namely SE-ISI (i.e., set extent and interval-set intent) formal concept, ISE-SI (i.e., interval-set extent and set intent) formal concept and ISE-ISI (i.e., interval-set extent and interval-set intent) formal concept. Although these three forms of partially-known formal concepts have been identified and proposed, their structures and relationships have not been fully investigated. The main objective of this paper is to provide such a study. We adopt a possible-world semantics of an incomplete formal context, i.e., an incomplete formal context is viewed as the family of all its possible completions. This enables us to systematically study the structures of the three different forms of partially-known formal concepts and their relationships. To be consistent with the possible-world semantics, we interpret a partially-known formal concept as the family of formal concepts in completions of an incomplete formal context. In addition to presenting theorems to summarize our results, we use an example to illustrate the main ideas.
![](http://media.springernature.com/m312/springer-static/image/art%3A10.1007%2Fs13042-017-0743-z/MediaObjects/13042_2017_743_Fig1_HTML.gif)
![](http://media.springernature.com/m312/springer-static/image/art%3A10.1007%2Fs13042-017-0743-z/MediaObjects/13042_2017_743_Fig2_HTML.gif)
![](http://media.springernature.com/m312/springer-static/image/art%3A10.1007%2Fs13042-017-0743-z/MediaObjects/13042_2017_743_Fig3_HTML.gif)
![](http://media.springernature.com/m312/springer-static/image/art%3A10.1007%2Fs13042-017-0743-z/MediaObjects/13042_2017_743_Fig4_HTML.gif)
![](http://media.springernature.com/m312/springer-static/image/art%3A10.1007%2Fs13042-017-0743-z/MediaObjects/13042_2017_743_Fig5_HTML.gif)
![](http://media.springernature.com/m312/springer-static/image/art%3A10.1007%2Fs13042-017-0743-z/MediaObjects/13042_2017_743_Fig6_HTML.gif)
Similar content being viewed by others
References
Belohlavek R (1998) Fuzzy concepts and conceptual structures: induced similarities. In: Joint Conf. Inf. Sci.’98 Proceedings, Vol I, pp 179-182. Durham, USA
Burmeister P (1987) ConImpa program for formal concept analysis. Technische Hochschule Darmstadt
Burmeister P, Holzer R (2000) On the treatment of incomplete knowledge in formal concept analysis. In: Proceedings of international conference on conceptual structures (ICCS 2000), pp 385-398. Darmstadt
Burusco A, Fuentes-Gonzalez R (1994) The study of the L-fuzzy concept lattice. Mathw Soft Comput 3:209–218
Djouadi Y, Dubois D, Prade P (2009) Différentes extensions floues de lanalyse formelle de concepts. Actes Renc. Franc. sur la Logique Floue et ses Applications Cépadues edn, pp 141-148. Toulouse
Djouadi Y, Dubois D, Prade P (2010) Graduality, uncertainty and typicality in formal concept analysis. In: Cornelis C, Deschrijver G, Nachtegael M, Schockaert S, Shi Y (eds) 35 years of fuzzy set theory. Springer, Heidelberg, pp 127–147
Dubois D, de Saint-Cyr FD, Prade H (2007) A possibility-theoretic view of formal concept analysis. Fundam Inform 75(1–4):195–213
Ganter B, Stumme G, Wille R (2005) Formal concept analysis: foundations and applications. Springer, Berlin, Heidelberg
Ganter B, Wille R (1999) Formal concept analysis: mathematical foundations. Springer, Berlin, Heidelberg
Holzer R (2004) Knowledge acquisition under incomplete knowledge using methods from formal concept analysis: part I and II. Fundam Inform 63(1):17–39 (41-63)
Huang CC, Li JH, Mei CL, Wu WZ (2017) Three-way concept learning based on cognitive operators: an information fusion viewpoint. Int J Approx Reason 83:218–242
Krupka M, Lastovicka J (2012) Concept lattices of incomplete data. In: Proceedings of international conference on formal concept analysis (ICFCA 2012): 180-194. Leuven
Li JH, Huang CC, Qi JJ, Qian YH, Liu WQ (2017) Three-way cognitive concept learning via multi-granularity. Inf Sci 378:244–263
Li JH, Kumar CA, Mei CL, Wang XZ (2017) Comparison of reduction in formal decision contexts. Int J Approx Reason 80:100–122
Li JH, Mei CL, Lv YJ (2013) Incomplete decision contexts: approximate concept construction, rule acquisition and knowledge reduction. Int J Approx Reason 54:149–165
Li JH, Mei CL, Xu WH, Qian YH (2015) Concept learning via granular computing: a cognitive viewpoint. Inf Sci 298:447–467
Li JH, Deng S (2017) Concept lattice, three-way decisions and their research outlooks. J Northwest Univ (Nat Sci Edition) 47(3):321–329
Li LF (2017) Multi-level interval-valued fuzzy concept lattices and their attribute reduction. Int J Mach Learn Cybern 8(1):45–56
Li Y, Zhang ZH, Chen WB (2017) TDUP: an approach to incremental mining of frequent itemsets with three-way-decision pattern updating. Int J Mach Learn Cybern 8(2):441–453
Lipski W Jr (1979) On semantic issues connected with incomplete information databases. ACM Trans Database Syst 4:269–296
Ma JM, Cai MJ, Zou CJ (2017) Concept acquisition approach of object-oriented concept lattices. Int J Mach Learn Cybern 8(1):123–134
Obiedkov S (2002) Modal logic for evaluating formulas in incomplete contexts. In: Proceedings of international conference on conceptual ctructures (ICCS 2002), pp. 314-325
Pollandt S (1997) Fuzzy begriffe. Springer, Berlin, Heidelberg
Qi JJ, Qian T, Wei L (2016) The connections between three-way and classical concept lattices. Knowl Based Syst 91:143–151
Qi JJ, Wei L, Yao YY (2014) Three-way formal concept analysis. In: Proceedings of RSKT 2014, LNCS (LNAI) vol 8818, pp 732-741
Ren RS, Wei L (2016) The attribute reductions of three-way concept lattices. Knowl Based Syst 99:92–102
Shao MW, Leung Y, Wang XZ, Wu WZ (2016) Granular reducts of formal fuzzy contexts. Knowl Based Syst 114:156–166
Shivhare R, Cherukuri AK (2017) Three-way conceptual approach for cognitive memory functionalities. Int J Mach Learn Cybern 8(1):21–34
Singh PK (2017) Three-way fuzzy concept lattice representation using neutrosophic set. Int J Mach Learn Cybern 8(1):69–79
Wille R (1982) Restructuring lattice theory: an approach based on hierarchies of concepts. In: Rival I (ed) Ordered Sets. Reidel Publishing Company, Dordrecht-Boston, pp 445–470
Yao YY (1993) Interval-set algebra for qualitative knowledge representation. In: Proceedings of the 5th international conference on computing and information, pp 370-374. Sudbury
Yao YY (2012) An outline of a theory of three-way decisions. In: Proceedings of RSCTC 2012, LNCS (LNAI) vol 7413, pp 1-17. Springer, Heidelberg
Yao YY (2017) Interval sets and three-way concept analysis in incomplete contexts. Int J Mach Learn Cybern 8(1):3–20
Yao YY (2016) Rough-set concept analysis: interpreting RS-definable concepts based on ideas from formal concept analysis. Inf Sci 346:442–462
Zhai JH, Zhang Y, Zhu HY (2017) Three-way decisions model based on tolerance rough fuzzy set. Int J Mach Learn Cybern 8(1):35–43
Zhao YX, Li JH, Liu WQ, Xu WH (2017) Cognitive concept learning from incomplete information. Int J Mach Learn Cybern 8(1):159–170
Acknowledgements
This work is partially supported by the National Natural Science Foundation of China (Grant Nos. 11371014 and 11671007), the State Scholarship Fund of China (Grant No. 201506970015), and a Discovery Grant from NSERC, Canada.
Author information
Authors and Affiliations
Corresponding author
Appendix: Proofs
Appendix: Proofs
Proof of Proposition 1
2. If \([\underline{O_{1}},\overline{O_{1}}]\preccurlyeq [\underline{O_{2}},\overline{O_{2}}]\), it follows that \(\underline{O_{1}}\subseteq \underline{O_{2}}\) and \(\overline{O_{1}}\subseteq \overline{O_{2}}\). Based on property (1) of the operator i, we obtain \(\underline{ i}(\underline{O_{2}})\subseteq \underline{ i}(\underline{O_{1}})\) and \(\overline{ i}(\overline{O_{2}})\subseteq \overline{ i}(\overline{O_{1}})\). Thus, \(\underline{ i}(\underline{O_{2}})\cap \overline{ i}(\overline{O_{2}})\subseteq \underline{ i}(\underline{O_{1}})\cap \overline{ i} (\overline{O_{1}})\). That is, \(\langle \underline{ i},\overline{ i} \rangle ([\underline{O_{2}},\overline{O_{2}}])\subseteq \langle \underline{ i},\overline{ i} \rangle ([\underline{O_{1}},\overline{O_{1}}])\). The part of attribute interval set can be proved similarly.
3. Based on property (2) of the operators i and e, we can get \(O\subseteq \underline{ e}(\underline{ i}(O))\) and \(O\subseteq \overline{ e}(\overline{ i}(O))\). Hence, \(O\subseteq \underline{ e}(\underline{ i}(O))\cap \overline{ e}(\overline{ i}(O)).\) That is, \(O\subseteq \langle \underline{ e},\overline{ e} \rangle ([\underline{ i},\overline{ i}](O))\). The second part can be proved similarly.
4. According to property (2) of the operators i and e, we obtain \(\underline{O_{1}}\subseteq \underline{ e}(\underline{ i}(O_{1}))\) and \(\overline{O_{1}}\subseteq \overline{ e}(\overline{ i}(\overline{O_{1}}))\). It is obvious that \(\underline{O_{1}}\subseteq \underline{ e}(\underline{ i}(\underline{O_{1}}))\cup \underline{ e}(\overline{ i}(\overline{O_{1}}))\) and \(\overline{O_{1}}\subseteq \overline{ e}(\underline{ i}(\underline{O_{1}}))\cup \overline{ e}(\overline{ i}(\overline{O_{1}}))\). Also, based on property (6) of the operator e, we can get \(\underline{ e}(\underline{ i}(\underline{O_{1}}))\cup \underline{ e}(\overline{ i}(\overline{O_{1}}))\subseteq \underline{ e}(\underline{ i}(\underline{O_{1}})\cap \overline{ i}(\overline{O_{1}}))\) and \(\overline{ e}(\underline{ i}(\underline{O_{1}}))\cup \overline{ e}(\overline{ i}(\overline{O_{1}}))\subseteq \overline{ e}(\underline{ i}(\underline{O_{1}})\cap \overline{ i}(\overline{O_{1}}))\). Thus, \(\underline{O_{1}}\subseteq \underline{ e}(\underline{ i}(\underline{O_{1}})\cap \overline{ i}(\overline{O_{1}}))\) and \(\overline{O_{1}}\subseteq \overline{ e}(\underline{ i}(\underline{O_{1}})\cap \overline{ i}(\overline{O_{1}}))\). Since \([\underline{ e},\overline{ e}](\langle \underline{ i},\overline{ i} \rangle ([\underline{O_{1}},\overline{O_{1}}]))=[\underline{ e}(\underline{ i}(\underline{O_{1}})\cap \overline{ i}(\overline{O_{1}})), \overline{ e}(\underline{ i}(\underline{O_{1}})\cap \overline{ i}(\overline{O_{1}}))]\), based on the definition of the partial order on interval set, we can get \([\underline{O_{1}},\overline{O_{1}}]\preccurlyeq [\underline{ e},\overline{ e}](\langle \underline{ i},\overline{ i} \rangle ([\underline{O_{1}},\overline{O_{1}}]))\). The rest part can be proved similarly.
5. By (4), we know \([\underline{ i},\overline{ i}](O)\preccurlyeq [\underline{ i},\overline{ i}](\langle \underline{ e},\overline{ e} \rangle ([\underline{ i},\overline{ i}](O)))\). Based on (1) and (3), it follows \([\underline{ i},\overline{ i}](\langle \underline{ e},\overline{ e} \rangle ([\underline{ i},\overline{ i}](O)))\preccurlyeq [\underline{ i},\overline{ i}](O)\). Thus \([\underline{ i},\overline{ i}](O)=[\underline{ i},\overline{ i}](\langle \underline{ e},\overline{ e} \rangle ([\underline{ i},\overline{ i}](O)))\). The rest part can be proved similarly.
6. By (3), we know \(\langle \underline{ i},\overline{ i}\rangle ([\underline{O_{1}},\overline{O_{1}}])\subseteq \langle \underline{ i},\overline{ i}\rangle ([ \underline{ e},\overline{ e}] (\langle \underline{ i},\overline{ i}\rangle ([\underline{O_{1}}\), \(\overline{O_{1}}])))\). Based on (2) and 4), it follows \(\langle \underline{ i},\overline{ i}\rangle ([ \underline{ e},\overline{ e}] (\langle \underline{ i},\overline{ i}\rangle ([\underline{O_{1}}\), \(\overline{O_{1}}])))\subseteq \langle \underline{ i},\overline{ i}\rangle ([\underline{O_{1}},\overline{O_{1}}])\). Thus \(\langle \underline{ i},\overline{ i}\rangle ([\underline{O_{1}},\overline{O_{1}}])=\langle \underline{ i},\overline{ i}\rangle ([ \underline{ e}\), \(\overline{ e}] (\langle \underline{ i},\overline{ i}\rangle ([\underline{O_{1}},\overline{O_{1}}])))\). The reset part can be proved similarly.
7. From the definition of the operator \([\underline{ i},\overline{ i}]\), we get \([\underline{ i},\overline{ i}]([\underline{O_{1}},\overline{O_{1}}])\sqcap [\underline{ i},\overline{ i}]([\underline{O_{2}},\overline{O_{2}}])=[\underline{ i}(\underline{O_{1}}),\overline{ i} (\overline{O_{1}})]\sqcap [\underline{ i}(\underline{O_{2}})\), \(\overline{ i}(\overline{O_{2}})]\). According to the intersection of interval sets, it follows \([\underline{ i}(\underline{O_{1}}),\overline{ i}(\overline{O_{1}})]\sqcap [\underline{ i}(\underline{O_{2}}),\overline{ i}(\overline{O_{2}})]= [\underline{ i}(\underline{O_{1}})\cap \underline{ i}(\underline{O_{2}}),\overline{ i}(\overline{O_{1}})\) \(\cap \overline{ i}(\overline{O_{2}})]\). By property (5) of operator i, we can get \([\underline{ i}(\underline{O_{1}})\cap \underline{ i}(\underline{O_{2}}),\overline{ i}(\overline{O_{1}}) \cap \overline{ i}(\overline{O_{2}})]= [\underline{ i}(\underline{O_{1}}\cup \underline{O_{2}}),\overline{ i}(\overline{O_{1}}\cup \overline{O_{2}})]\). Because of the definition of the operator \([\underline{ i},\overline{ i}]\), we have \([\underline{ i}(\underline{O_{1}}\cup \underline{O_{2}}),\overline{ i}(\overline{O_{1}}\cup \overline{O_{2}})]=[\underline{ i},\overline{ i}]([\underline{O_{1}}\cup \underline{O_{2}},\overline{O_{1}}\cup \overline{O_{2}}])\). According to the union of interval sets, it follows \([\underline{ i},\overline{ i}]([\underline{O_{1}}\cup \underline{O_{2}},\overline{O_{1}}\cup \overline{O_{2}}])= [\underline{ i},\overline{ i}]([\underline{O_{1}},\overline{O_{1}}]\sqcup [\underline{O_{2}},\overline{O_{2}}])\). Thus \([\underline{ i},\overline{ i}]([\underline{O_{1}},\overline{O_{1}}])\sqcap [\underline{ i},\overline{ i}]([\underline{O_{2}},\overline{O_{2}}])= [\underline{ i},\overline{ i}]([\underline{O_{1}},\overline{O_{1}}]\sqcup [\underline{O_{2}},\overline{O_{2}}])\). The rest part can be proved similarly.
8. According to the union of interval sets, it follows \(\langle \underline{ i},\overline{ i}\rangle ([\underline{O_{1}},\overline{O_{1}}]\sqcup [\underline{O_{2}},\overline{O_{2}}])=\langle \underline{ i},\overline{ i}\rangle ([\underline{O_{1}}\cup \underline{O_{2}}, \overline{O_{1}}\cup \overline{O_{2}}])\). Because of the definition of operator \(\langle \underline{ i},\overline{ i}\rangle\), we have \(\langle \underline{ i},\overline{ i}\rangle ([\underline{O_{1}}\cup \underline{O_{2}}, \overline{O_{1}}\cup \overline{O_{2}}])=\underline{ i}(\underline{O_{1}}\cup \underline{O_{2}})\cap \overline{ i}(\overline{O_{1}} \cup \overline{O_{2}})\). By property 5) of the operator i, it follows \(\underline{ i}(\underline{O_{1}}\cup \underline{O_{2}})\cap \overline{ i}(\overline{O_{1}}\cup \overline{O_{2}})= \underline{ i}(\underline{O_{1}})\cap \underline{ i}(\underline{O_{2}})\cap (\overline{ i}(\overline{O_{1}}) \cap \overline{ i}(\overline{O_{2}}))= \underline{ i}(\underline{O_{1}})\cap (\overline{ i}(\overline{O_{1}}))\cap (\underline{ i}(\underline{O_{2}}) \cap \overline{ i}(\overline{O_{2}}))\). Based on the definition of operator \(\langle \underline{ i},\overline{ i}\rangle\), it follows \((\underline{ i}(\underline{O_{1}})\cap (\overline{ i}(\overline{O_{1}}))\cap (\underline{ i}(\underline{O_{2}}) \cap \overline{ i}(\overline{O_{2}}))= \langle \underline{ i},\overline{ i}\rangle ([\underline{O_{1}},\overline{O_{1}}])\cap \langle \underline{ i},\overline{ i}\rangle ([\underline{O_{2}},\overline{O_{2}}])\). Thus \(\langle \underline{ i},\overline{ i}\rangle ([\underline{O_{1}},\overline{O_{1}}]\sqcup [\underline{O_{2}},\overline{O_{2}}])=\langle \underline{ i},\overline{ i}\rangle ([\underline{O_{1}},\overline{O_{1}}])\cap \langle \underline{ i},\overline{ i}\rangle ([\underline{O_{2}},\overline{O_{2}}])\). The rest part can be proved similarly. \(\square\)
Proof of Theorem 3
(A) We only prove the infimum part and the supermum part can be proved dually. Now, we prove that \((O_{1}\cap O_{2}, [\underline{ i},\underline{ i}](\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\) is an SE-ISI concept. By 8) of Proposition 1, it follows that \(\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}]\) \(\sqcup [\underline{A_{2}},\overline{A_{2}}])=\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}])\cap \langle \underline{ e},\overline{ e}\rangle ([\underline{A_{2}},\overline{A_{2}}])\). Since \((O_{1}\), \([\underline{A_{1}},\overline{A_{1}}])\) and \((O_{2},[\underline{A_{2}},\overline{A_{2}}])\) are SE-ISI concepts, it follows that \(\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}])=O_{1}\) and \(\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{2}},\overline{A_{2}}])=O_{2}\). Thus \(\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}])\cap \langle \underline{ e},\overline{ e}\rangle ([\underline{A_{2}},\overline{A_{2}}])=O_{1}\cap O_{2}\). Hence, we can get \([\underline{ i},\overline{ i}](O_{1}\cap O_{2})=[\underline{ i},\overline{ i}](\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}]))\). Then by 6) of Proposition 1, it follows that \(\langle \underline{ e},\overline{ e}\rangle ([\underline{ i},\overline{ i}](\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))=\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])\). According to 8) of Proposition 1, we have \(\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])=\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}])\cap \langle \underline{ e},\overline{ e}\rangle ([\underline{A_{2}}\), \(\overline{A_{2}}])\). Since, \((O_{1},[\underline{A_{1}},\overline{A_{1}}])\) and \((O_{2},[\underline{A_{2}},\overline{A_{2}}])\) are SE-ISI concepts, it follows that \(\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}])=O_{1}\) and \(\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{2}},\overline{A_{2}}])=O_{2}\). Thus \(\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}])\cap \langle \underline{ e},\overline{ e}\rangle ([\underline{A_{2}}\), \(\overline{A_{2}}])=O_{1}\cap O_{2}\). Hence, we can get \(\langle \underline{ e},\overline{ e}\rangle ([\underline{ i},\overline{ i}](\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}}\), \(\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))=O_{1}\cap O_{2}\).
Then, we prove that \((O_{1}\cap O_{2}, [\underline{ i},\overline{ i}](\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\) is the infimum of SE-ISI concepts \((O_{1},[\underline{A_{1}}\), \(\overline{A_{1}}])\) and \((O_{2},[\underline{A_{2}},\overline{A_{2}}])\). Since \(O_{1}\cap O_{2}\subseteq O_{1}\) and \(O_{1}\cap O_{2}\subseteq O_{2}\), \((O_{1}\cap O_{2}, [\underline{ i},\overline{ i}](\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\preccurlyeq _\mathrm{{SE-ISI}}(O_{1},[\underline{A_{1}},\overline{A_{1}}])\) and \((O_{1}\cap O_{2}, [\underline{ i},\overline{ i}](\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\) \(\preccurlyeq _\mathrm{{SE-ISI}}(O_{2},[\underline{A_{2}},\overline{A_{2}}])\) hold. Hence \((O_{1}\cap O_{2}, [\underline{ i},\overline{ i}](\langle \underline{ e},\overline{ e}\rangle\) \(([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\) is the lower bound of two SE-ISI concepts \({(O_{1},[\underline{A_{1}},\overline{A_{1}}])}\) and \({(O_{2},[\underline{A_{2}},\overline{A_{2}}])}\). If there is any SE-ISI concept \((O_{i},[\underline{A_{i}},\overline{A_{i}}])\) that is the lower bound of \((O_{1},[\underline{A_{1}},\overline{A_{1}}])\) and \((O_{2},[\underline{A_{2}},\overline{A_{2}}])\), then the formulas \((O_{i},[\underline{A_{i}},\overline{A_{i}}])\preccurlyeq _\mathrm{{SE-ISI}}(O_{1},[\underline{A_{1}},\overline{A_{1}}])\) and \((O_{i},[\underline{A_{i}}\), \(\overline{A_{i}}])\preccurlyeq _\mathrm{{SE-ISI}}(O_{2},[\underline{A_{2}},\overline{A_{2}}])\) hold. Thus \(O_{i}\subseteq O_{1}\) and \(O_{i}\subseteq O_{2}\) hold. Hence \(O_{i}\subseteq O_{1}\cap O_{2}\). That is, \((O_{i},[\underline{A_{i}},\overline{A_{i}}])\) \(\preccurlyeq _\mathrm{{SE-ISI}}(O_{1}\cap O_{2}, [\underline{ i},\overline{ i}](\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\). Thus \((O_{1}\cap O_{2}, [\underline{ i},\overline{ i}](\langle \underline{ e},\overline{ e}\rangle ([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\) is the infimum of SE-ISI concepts \((O_{1},[\underline{A_{1}},\overline{A_{1}}])\) and \((O_{2},[\underline{A_{2}},\overline{A_{2}}])\).
(B) Since ISE-SI concepts and SE-ISI concepts are dual to each other. The results of (B) can be proved dually to (A).
(C) We only prove the infimum part and the supermum part can be proved dually. Now, we prove that \(([\underline{O_{1}},\overline{O_{1}}]\sqcap [\underline{O_{2}},\overline{O_{2}}],[\underline{ i},\overline{ i}]([\underline{ e},\overline{ e}]([\underline{A_{1}},\) \(\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\) is an ISE-ISI concept. Since \(([\underline{O_{1}}, \overline{O_{1}}],\) \([\underline{A_{1}}, \overline{A_{1}}])\) and \(([\underline{O_{2}}, \overline{O_{2}}], [\underline{A_{2}}, \overline{A_{2}}])\) are ISE-ISI concepts, it is obvious that \((\underline{O_{1}},\underline{A_{1}})\) and \((\underline{O_{2}},\underline{A_{2}})\) are formal concepts in least completion \({\mathbb {K}}_{*}\). Thus according to the way of computing the infimum of formal concepts in a complete formal context, \((\underline{O_{1}}\cap \underline{O_{2}},\underline{ i}(\underline{ e}(\underline{A_{1}}\cup \underline{A_{2}})))\) is the infimum of formal concepts \((\underline{O_{1}},\underline{A_{1}})\) and \((\underline{O_{2}},\underline{A_{2}})\). Hence \((\underline{O_{1}}\cap \underline{O_{2}},\underline{ i}(\underline{ e}(\underline{A_{1}}\cup \underline{A_{2}})))\) is a formal concept in context \({\mathbb {K}}_{*}\). Similarly, since \(([\underline{O_{1}}, \overline{O_{1}}], [\underline{A_{1}}, \overline{A_{1}}])\) and \(([\underline{O_{2}}, \overline{O_{2}}],\) \([\underline{A_{2}}, \overline{A_{2}}])\) are ISE-ISI formal concepts, it is obvious that \((\overline{O_{1}},\overline{A_{1}})\) and \((\overline{O_{2}},\overline{A_{2}})\) are formal concepts in greatest completion \({\mathbb {K}}^{*}\). Thus according to the way of computing the infimum of formal concepts in a complete formal context, \((\overline{O_{1}}\cap \overline{O_{2}},\overline{ i}(\overline{ e}(\overline{A_{1}}\cup \overline{A_{2}})))\) is the infimum of concepts \((\overline{O_{1}},\overline{A_{1}})\) and \((\overline{O_{2}},\overline{A_{2}})\). Hence \((\overline{O_{1}}\cap \overline{O_{2}},\overline{ i}(\overline{ e}(\overline{A_{1}}\cup \overline{A_{2}})))\) is a formal concept in context \({\mathbb {K}}^{*}\). Also, since \(\underline{O_{1}}\subseteq \overline{O_{1}}\) and \(\underline{O_{2}}\subseteq \overline{O_{2}}\), we can get \(\underline{O_{1}}\cap \underline{O_{2}}\subseteq \overline{O_{1}}\cap \overline{O_{2}}\). Thus \(([\underline{O_{1}}\cap \underline{O_{2}},\overline{O_{1}}\cap \overline{O_{2}}],[\underline{ i}(\underline{ e}(\underline{A_{1}}\cup \underline{A_{2}})),\overline{ i}(\overline{ e}(\overline{A_{1}}\cup \overline{A_{2}}))])\) is an ISE-ISI concept. That is, \(([\underline{O_{1}},\overline{O_{1}}]\sqcap [\underline{O_{2}},\overline{O_{2}}],[\underline{ i},\overline{ i}]([\underline{ e},\overline{ e}]([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\) is an ISE-ISI concept.
Then, we prove that \(([\underline{O_{1}},\overline{O_{1}}]\sqcap [\underline{O_{2}},\overline{O_{2}}],[\underline{ i},\overline{ i}]([\underline{ e},\overline{ e}]\) \(([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\) is the infimum of ISE-ISI concepts \(([\underline{O_{1}}, \overline{O_{1}}], [\underline{A_{1}}, \overline{A_{1}}])\) and \(([\underline{O_{2}}, \overline{O_{2}}], [\underline{A_{2}}, \overline{A_{2}}])\). Since the formulas \([\underline{O_{1}},\overline{O_{1}}]\sqcap [\underline{O_{2}},\overline{O_{2}}]\preccurlyeq [\underline{O_{1}},\overline{O_{1}}]\) and \([\underline{O_{1}},\overline{O_{1}}]\sqcap [\underline{O_{2}},\overline{O_{2}}]\preccurlyeq [\underline{O_{2}},\overline{O_{2}}]\) hold, it is obviously that \(([\underline{O_{1}},\overline{O_{1}}]\sqcap [\underline{O_{2}},\overline{O_{2}}],[\underline{ i},\overline{ i}]([\underline{ e},\overline{ e}]([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\) is the lower bound of ISE-ISI concepts \(([\underline{O_{1}}, \overline{O_{1}}], [\underline{A_{1}}, \overline{A_{1}}])\) and \(([\underline{O_{2}}, \overline{O_{2}}],\) \([\underline{A_{2}}, \overline{A_{2}}])\). If there is any ISE-ISI concept \(([\underline{O_{i}},\overline{O_{i}}],[\underline{A_{i}},\) \(\overline{A_{i}}])\) that is the lower bound of ISE-ISI concepts \(([\underline{O_{1}}, \overline{O_{1}}],\) \([\underline{A_{1}}, \overline{A_{1}}])\) and \(([\underline{O_{2}}, \overline{O_{2}}], [\underline{A_{2}}, \overline{A_{2}}])\), the formula \([\underline{O_{i}},\overline{O_{i}}]\preccurlyeq [\underline{O_{1}}, \overline{O_{1}}]\) and \([\underline{O_{i}},\overline{O_{i}}]\preccurlyeq [\underline{O_{2}}, \overline{O_{2}}]\) hold. Thus \([\underline{O_{i}},\overline{O_{i}}]\preccurlyeq [\underline{O_{1}}, \overline{O_{1}}]\sqcap [\underline{O_{2}}, \overline{O_{2}}]\). That is, \(([\underline{O_{i}},\overline{O_{i}}],[\underline{A_{i}},\overline{A_{i}}])\preccurlyeq _\mathrm{{ISE-ISI}}([\underline{O_{1}},\overline{O_{1}}]\sqcap [\underline{O_{2}},\overline{O_{2}}],[\underline{ i},\overline{ i}]([\underline{ e},\overline{ e}]([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\). Thus \(([\underline{O_{1}},\overline{O_{1}}]\sqcap [\underline{O_{2}},\overline{O_{2}}],[\underline{ i},\overline{ i}]([\underline{ e},\overline{ e}]([\underline{A_{1}},\overline{A_{1}}]\sqcup [\underline{A_{2}},\overline{A_{2}}])))\) is the infimum of ISE-ISI concepts \(([\underline{O_{1}}, \overline{O_{1}}], [\underline{A_{1}}, \overline{A_{1}}])\) and \(([\underline{O_{2}}, \overline{O_{2}}], [\underline{A_{2}}, \overline{A_{2}}])\). \(\square\)
Proof of Theorem 4
Sufficiency. First, we prove that if \((O,[A_{1},A_{2}])\) is an SE-ISI concept and the condition \(O=\underline{ e}(A_{1})\) holds, then there exists an ISE-ISI concept \(([\underline{ e}(A_{1}),\overline{ e}(A_{2})],[A_{1},A_{2}])\) having the same intent as \((O,[A_{1},A_{2}])\). If \((O,[A_{1},A_{2}])\) is an SE-ISI concept, then we have \(\underline{ i}(O)=A_{1}\), \(\overline{ i}(O)=A_{2}\) and \(\underline{ e}(A_{1})\bigcap \overline{ e}(A_{2})=O\). Thus we obtain \(O\subseteq \underline{ e}(A_{1})\) and \(O\subseteq \overline{ e}(A_{2})\). Hence \(\underline{ i}(\underline{ e}(A_{1}))\subseteq \underline{ i}(O)\) and \(\overline{ i}(\overline{ e}(A_{2}))\subseteq \overline{ i}(O)\). That is, \(\underline{ i}(\underline{ e}(A_{1}))\subseteq A_{1}\) and \(\overline{ i}(\overline{ e}(A_{2}))\subseteq A_{2}\). Because we also know \(A_{1}\subseteq \underline{ i}(\underline{ e}(A_{1}))\) and \(A_{2}\subseteq \overline{ i}(\overline{ e}(A_{2}))\), we can get \(A_{1}=\underline{ i}(\underline{ e}(A_{1}))\) and \(A_{2}=\overline{ i}(\overline{ e}(A_{2}))\). Thus \((\underline{ e}(A_{1}),A_{1})\) is an SE-SI concept in the least completion \({\mathbb {K}}_{*}\) and \((\overline{ e}(A_{2}),A_{2})\) is an SE-SI concept in the greatest completion \({\mathbb {K}}^{*}\). Since \(O=\underline{ e}(A_{1})\bigcap \overline{ e}(A_{2})\) and the condition \(O=\underline{ e}(A_{1})\) holds, it is obviously \(\underline{ e}(A_{1})=O\subseteq \overline{ e}(A_{2})\). Hence \(([\underline{ e}(A_{1}),\overline{ e}(A_{2})],[A_{1},A_{2}])\) is an ISE-ISI concept.
Then, we prove that if \(([O_{1},O_{2}],[A_{1},A_{2}])\) is an ISE-ISI concept and the condition \(\overline{ i}(O_{1})=A_{2}\) holds, there exists an SE-ISI concept \((O_{1},[A_{1},A_{2}])\) having the same intent as \(([O_{1},O_{2}],[A_{1},A_{2}])\). Since \(([O_{1}, O_{2}],[A_{1}, A_{2}])\) is an ISE-ISI concept, we have \(\underline{ i}(O_{1})=A_{1}\), \(\underline{ e}(A_{1})=O_{1}\), \(\overline{ e}(A_{2})=O_{2}\) and \(O_{1}\subseteq O_{2}\). Thus \(\underline{ e}(A_{1})\bigcap \overline{ e}(A_{2})=O_{1}\bigcap O_{2}=O_{1}\). Also, since the condition \(\overline{ i}(O_{1})=A_{2}\) holds, according to the definition of SE-ISI concept, \((O_{1},[A_{1},A_{2}])\) is an SE-ISI concept.
Necessity. Suppose \(O\ne \underline{ e}(A_{1})\). Since \(([\underline{ e}(A_{1}),\overline{ e}(A_{2})],\) \([A_{1},A_{2}])\) is an ISE-ISI concept, we have \(\underline{ e}(A_{1})\subseteq \overline{ e}(A_{2})\). Hence \(\underline{ e}(A_{1})\bigcap \overline{ e}(A_{2})=\underline{ e}(A_{1})\ne O\) which contradicts the condition that \((O,[A_{1},A_{2}])\) is an SE-ISI concept. Thus we have \(O=\underline{ e}(A_{1})\). If \((O,[A_{1}, A_{2}])\) is an SE-ISI concept, from the definition of SE-ISI concept, it is easy to get \(\overline{i}(O)=A_{2}.\) Because we have proved \(O=\underline{ e}(A_{1})=O_{1}\), we can get \(\overline{i}(O_{1})=A_{2}\). \(\square\)
Proof of Theorem 7
Since \(([\underline{O},\overline{O}],[\underline{A},\overline{A}])=\{(O,A)|O\in [\underline{O},\overline{O}], A\in [\underline{A},\overline{A}]\}\) is an ISE-ISI concept, we have that \((\underline{O},\underline{A})\) is the lower bound of \(([\underline{O},\overline{O}],[\underline{A},\overline{A}])\) and it is also an SE-SI concept in least completion \(\mathbb {K_{*}}\). Because of \((O_{1},A_{1})\in ([\underline{O},\overline{O}],[\underline{A},\overline{A}])\), we know \(\underline{O}\subseteq O_{1}\) and \(\underline{A}\subseteq A_{1}\). Since \((O_{1},A_{1})\) is also an SE-SI concept of least completion \(\mathbb {K_{*}}\), based on the property of the SE-SI concept, if \(\underline{O}\subseteq O_{1}\) holds, then we can obtain \(A_{1}\subseteq \underline{A}\), and if \(\underline{A}\subseteq A_{1}\) holds, then we can get \(O_{1}\subseteq \underline{O}\). Thus \(A_{1}=\underline{A}\) and \(O_{1}=\underline{O}\). That is, \((O_{1},A_{1})=(\underline{O},\underline{A})\). The last part of this theorem can be proved similarly. \(\square\)
Proof of Theorem 8
Necessity. If (O, A) is an SE-SI concept in completion \(\mathbb {K}\), then we have \(e_{\mathbb {K}}(A)=O\). Since \(\underline{ e}(A)\subseteq e_{\mathbb {K}}(A)=O\), we obtain \(\underline{ e}(A)\subseteq O\).
Sufficiency. We show that this is indeed the case by constructing a completion \(\mathbb {K} =(O\!B, AT, \mathbf{I_{\mathbb {K}}})\) as \(\mathbf{I_{\mathbb {K}}}=\mathbf{I_*} \cup (O \times A)\), where \(\mathbf{I_*}\) is the binary relation of the least completion \(\mathbb {K_{*}}\). The completion \(\mathbb {K}\) is obtained by changing all ? into \(+\) for object-attribute pairs in \(O\times A\) and changing all ? into − for all object-attribute pairs in \((O\!B - O)\times (AT - A)\). It is obvious that \(O\subseteq e_{\mathbb {K}}(A)\).
In the following, we prove that \(O= e_{\mathbb {K}}(A)\). Suppose \(O\subset e_{\mathbb {K}}(A)\), there must exist an object \(o\in e_{\mathbb {K}}(A)\setminus O\). Thus we have \(o\in \underline{ e}(A)\) that contradicts the condition \(\underline{ e}(A)\subseteq O\). Hence the equation \(O= e_{\mathbb {K}}(A)\) holds.
Now we prove \(i_{\mathbb {K}}(O)=A\). Based on the definition of the completion \(\mathbb {K}\), \(A\subseteq i_{\mathbb {K}}(O)\). Suppose \(A\subset i_{\mathbb {K}}(O)\), there must exist an attribute \(a\in i_{\mathbb {K}}(O)\setminus A\). Thus we have \(a\in \underline{ i}(O)=\underline{A}\) that contradicts the condition \(\underline{A}\subseteq A\). Hence \(i_{\mathbb {K}}(O)=A\) holds. Thus we obtain that (O, A) is an SE-SI concept in completion \(\mathbb {K}\). \(\square\)
Proof of Theorem 9
If (O, A) is an SE-SI concept in the least completion \(\mathbb {K_{*}}\), we have \(\underline{ i}(O)=A\) and \(\underline{ e}(A)=O\). Since \(O\subseteq \overline{ e}(\overline{ i}(O))\) holds, we obtain \(\underline{ e}(A)\bigcap \overline{ e}(\overline{ i}(O))=O\bigcap \overline{ e}(\overline{ i}(O))=O\). Thus \((O,(A,\overline{ i}(O)))\) is an SE-ISI concept in the incomplete context \(\mathbb {IK}\). Similarly, the later part can be proved. \(\square\)
Rights and permissions
About this article
Cite this article
Ren, R., Wei, L. & Yao, Y. An analysis of three types of partially-known formal concepts. Int. J. Mach. Learn. & Cyber. 9, 1767–1783 (2018). https://doi.org/10.1007/s13042-017-0743-z
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s13042-017-0743-z